Below is the unedited draft of:
The precis of: Structure, Function, and Dynamics: An Integrated Approach to Neural Organization, By Michael Arbib,  Péter Érdi and John Szentagothai. (1997) Behavioral and Brain Sciences 23 (4): XXX-XXX.

For information about subscribing or purchasing offprints of the published version, with commentaries and author's response, write to: journals_subscriptions@cup.org (North America) or journals_marketing@cup.cam.ac.uk (All other countries).

Précis of "Structure, Function, and Dynamics: An Integrated Approach to Neural Organization" for BBS multiple book review

Structure, Function and Dynamics : An Integrated Approarch to Neural Organization was published by MIT  Press in 1997.

Michael Arbib
Director, USC Brain Project,
University of Southern California,
Los Angeles, CA 90089-2520, USA.

Arbib@pollux.usc.edu
http://www-hbp.usc.edu/

Péter Érdi
Head, Dept. Biophysics
KFKI Research Institute for Particle and Nuclear Physics of the Hungarian Academy of Sciences
H-1525 Budapest, P.O. Box 49, Hungary.

erdi@rmki.kfki.hu
http://www.rmki.kfki.hu/biofiz/biophysics.html

Michael A. Arbib is Professor of Computer Science (of which he is also Chair), Neuroscience, Biomedical Engineering, Electrical Engineering, and Psychology at the University of Southern California (USC). Born in England, Arbib grew up in Australia and received his Ph.D. in Mathematics from MIT. After five years at Stanford, he became chairman of Computer and Information Science at the University of Massachusetts, Amherst in 1970, and remained in the Department until moving to USC in 1986. The author or editor of more than 30 books, Arbib recently edited The Handbook of Brain Theory and Neural Networks. His current research focuses on brain mechanisms of visuomotor behavior, on neuroinformatics, and on the evolution of language.

Péter Érdi is a Head of the Dept. Biophysics of the KFKI Research Institute for Particle and Nuclear Physics of the Hungarian Academy of Sciences. He is Professor of Technical University of Budapest, and Kossuth University Debrecen, and has a Szechenyi Professor Fellowship (1999-2002) to the EÖtvÖs University, Budapest. He has been working on the the applications of the theory of nonlinear dynamic systems to chemical and biological phenomena. The activity of his computational neuroscience research group has focused on the theoretical and computational approach to the olfactory system and of the hippocampus. Érdi has been a close coworker of John Szentagothai between 1981 and 1994.

Abstract

Neural Organization: Structure, Function, and Dynamics (Arbib, Érdi, and Szentágothai, 1997, Cambridge, MA: The MIT Press; henceforth Organization) shows how theory and experiment can supplement each other in an integrated, evolving account of structure, function, and dynamics. New data lead to new models; new models suggest the design of new experiments. Much of modern neuroscience seems excessively reductionist, focusing on the study of ever smaller microsystems with little appreciation of their contribution to the behaving organism. We welcome these new data but are concerned to restore some equilibrium between systems, cellular, and molecular neuroscience. After a brief tribute to our late colleague John Szentágothai, we trace the threads of Structure, Function and Dynamics as they weave through the book, thus providing a broad general framework for the integration of computational and empirical neuroscience. Part II of Organization presents a structural analysis of various brain regions – olfactory bulb and cortex, hippocampus, cerebral cortex, cerebellum, and basal ganglia - as prelude to our account of the dynamics of the neural circuits and function of each region. To exemplify this approach, this précis analyzes the hippocampus in anatomical, dynamical, and functional terms. We conclude by pointing the way to the use of our methodology in the development of Cognitive Neuroscience.

Keywords: neural organization, dynamics, Szentgothai, computational neuroscience, neural modeling, modular architectonics, neural plasticity, hippocampus, rhythmogenesis, cognitive maps, memory.

P1. In Memory of John Szentágothai

The present précis and BBS multiple book review is dedicated, with respect and affection, to the memory of János (John) Szentágothai. It is based on the book Neural Organization: Structure, Function, and Dynamics (Arbib, Érdi, and Szentágothai, 1997; we refer to the book as Organization in what follows).

The idea of writing Organization arose when the three authors took part in the first week of a School organized by Francesco Ventriglia on "Neural Modeling and Neural Networks" which was held on the Isle of Capri in October of 1992, a week which included the celebration of John Szentágothai’s eightieth birthday. Szentágothai and Arbib had previously co-authored Conceptual Models of Neural Organization (MIT Press 1975) and Szentágothai and Érdi had written papers together on the self-organization of the nervous system. These collaborations provided the basis for the new book. A lengthy draft of the book had been completed at the time of John Szentágothai’s death in September of 1994. In fact, John was working on the book that very morning. We thus start with a few words of appreciation for John's career.

John Szentágothai (1912-1994) is known for his many pioneering contributions to neuroanatomy. His scientific career began in the mid-thirties, when he helped verify the neuron doctrine against the reticular theory. (His early papers appeared under his original family name of Schimert.) In the late thirties and early forties he elaborated his secondary generation method as a technique for detecting pathways between brain regions. Szentágothai served in the Chair of Anatomy at Pécs University Medical School from 1946 to 1963. Combining anatomical and physiological methods, he made pioneering studies on the vestibulo-ocular reflex arc, then worked on the functional anatomy of spinal cord, brainstem and cerebellum. He was also involved in neuroembryological and neuroendocrinological research. In 1963, Szentágothai moved to the 1st Department of Anatomy of the Semmelweis University Medical School, Budapest, where he worked till the penultimate day of his life.

Szentágothai’s anatomical discoveries in the cerebellum, together with the physiological findings of John Eccles and Masao Ito, led to a fruitful cooperation and an epoch-making monograph on The Cerebellum as a Neuronal Machine (Springer-Verlag, 1967). From the late sixties his research concentrated on the functional organization of the cerebral cortex. He formulated (and refined in the light of new data) the modular architectonic principle of the cerebral cortex as the anatomical basis for physiologically defined cortical modules. He searched for "the essence of the neural" and hoped to find it in the self-organization of spontaneous (random) activity into biologically significant spatiotemporal activity. A very characteristic autobiography entitled "Too 'much' and too 'soon'" was written for a Festschrift dedicated for his seventieth (Szentágothai, 1982). For a brief summary of his activity written for his eightieth birthday, see Záborszky et al. (1992).

Szentágothai’s final reflections on neuroscience are preserved in Organization, especially in Chapter 2 (save for the last section), Chapter 7 (save for the last section), Section 8.1, Section 9.1, and much of Section 9.2. We thank Tamás Freund, Attila Gulyás, Miklos Réthelyi, George Székely and especially Jozsef Takács for their help in editing portions of this material. In some sense, this material - as Szentágothai’s last scientific writings - became a "sacred text" which we could not change beyond minor editing, much though we would have enjoyed debating some of his ideas with him had John still been alive. As we completed the other sections, our continuing "conversations" with John, based on many earlier interactions, strongly influenced our work.

P2. Introduction

Organization provides a comprehensive view of neural organization in the spirit of the cooperative development of theory and experiment. A "good" model is responsive to available data; an "interesting" set of data will test hypotheses which are theory-laden, whether the theory be formal or not. Our task is not to provide "final models" or "a complete unified theory of the brain". Rather, we seek to show how theory and experiment can supplement each other in an integrated, evolving account of structure, function, and dynamics. Much of modern neuroscience seems to us excessively reductionist, focusing on the study of ever smaller microsystems to the exclusion of an appreciation of their contribution to the behaving organism. We do not reject the data gained in this way, but are concerned to restore some equilibrium between systems neuroscience, cellular neuroscience, and molecular neuroscience.

For example, one of Organization's recurrent themes is to bridge different levels of organization by linking the learning rules which structure a variety of brain regions both to the functional roles of those regions and to the emerging understanding of the neurochemistry of synaptic plasticity and its variation from region to region. This is but one of many ways in which we exemplify how theory and experiment may be intricately intertwined in a continuing cycle of analysis and synthesis. We now turn to a brief characterization of the three approaches - structural, functional, and dynamical - which inform our account of neural organization. The obvious identification of authors is Szentágothai <-> Structure, Arbib <-> Function, and Érdi <-> Dynamics, but this is only a first approximation, for a functional model may involve more or less dynamics, and vice versa, and the anatomical data necessary to ground a model may be more or less those which attracted Szentágothai’s attention.

P2.1. Structural Approach

Studies of brain function and dynamics build on, and contribute to, an understanding of many brain regions and of the neural circuits which constitute them. Organization thus reviews anatomical data that integrate the overall spatial relations between a variety of brain regions with a selection of critical details of neural morphology and synaptic connectivity. This analysis of neural structure is guided by a developmental view which approaches the complexity of the adult nervous system through an understanding of the way in which that complexity emerges during embryogenesis, thus linking the structural approach to dynamical models of self-organization. The developing nervous system can generate movement before it becomes responsive to sensory stimuli, consonant with the emphasis on action-oriented perception in our functional studies, analyzing the ways in which sensory systems are specialized to serve a variety of behaviors. As a basis for our functional and dynamical analysis of a variety of systems, later chapters progress through regions of the brain which, singly or in combination, underlie these systems: the segmented part of the neuraxis (discussed as a case study in the Structural Overview of Chapter 2), the olfactory system, the hippocampus, the thalamus, the cerebral cortex, the cerebellum, and, finally, the basal ganglia.

P2.2. Functional Approach

Organization first approaches complex functions such as the control of eye movements, reaching and grasping, the use of a cognitive map for navigation, and the roles of vision in these behaviors, by the use of schemas in the sense of units which provide a functional decomposition of the overall skill or behavior. A schema account becomes a brain model when we offer hypotheses as to how each schema is implemented through the interaction of specific brain regions. A brain-based schema model may be tested by analysis of the behavior of animals with localized lesions or reversible inactivation of specific brain regions or by human brain imaging. Such a model provides the basis for modeling the overall function by neural networks which plausibly implement (usually in a distributed fashion) the schemas in the brain. Further analysis may then proceed bottom up (as the neural data drive further research) as well as top down (as we refine our schema-theoretic formulations).

The models in Organization use neural networks in the sense of computational neuroscience in which the structure of the network and the function of the neuron is constrained, at some appropriate level of detail, by the data of neuroanatomy and neurophysiology. This is in contrast to neural networks in the sense of connectionism in which the structure of the network is generic (e.g., a multilayered feedforward network, or a fully connected network) and the connections are determined by some "learning rule" which may be non-biological, as in the case of backpropagation, rather than constrained by anatomical data.

P2.3. Dynamical Approach

Dynamic system theory offers a conceptual and mathematical framework to analyze spatiotemporal neural phenomena occurring at different levels of organization, such as oscillatory and chaotic activity both in single neurons and in (often synchronized) neural networks, the self-organizing development and plasticity of ordered neural structures, and learning and memory phenomena associated with synaptic modification. We discuss a variety of rhythms (and arhythmia) found in the olfactory bulb and olfactory cortex, in the hippocampus, and in the thalamocortical system. In most cases we relate these rhythms to memory functions. We also study learning rules in both developmental processes (self-organization) and in the acquisition of a variety of behaviors. In this way, Organization grounds our functional analysis of neural organization in a dynamic systems analysis of the neural networks which implement the basic schemas.

P3. An Outline of Organization and This Article

Part I of Organization, "Overviews", opens with a Chapter, "The Many Themes of Neural Organization," which expands upon the above discussion of Structure, Dynamics, and Function to introduce a variety of themes which weave in and out of subsequent Chapters, binding Organization into a moderately coherent whole. We then devote three Chapters to detailed overviews of our three methods for understanding neural organization: a Structural Overview, a Functional Overview and a Dynamical Overview.

Part II, "Interacting Systems of the Brain," uses a structural organization to order our integrated approach to structure, function, and dynamics. Almost all of Chapters 5 through 10 begin with a structural analysis of a specific brain region as the prelude to our account of the dynamics of the neural circuits and of the function of the region. Learning, memory and plasticity are discussed in this functional and dynamic context. Chapter 5 looks at the role of rhythm generation and chaotic patterns in both olfactory bulb and olfactory cortex. Chapter 6 analyzes rhythm generation in, and memory functions of, hippocampus, as well as providing an extensive account of cognitive maps in the rat and declarative memory in humans. Chapter 7 offers a primarily structural account of the thalamus which emphasizes that, far from being simply a set of relay structures, it binds the cerebral cortex in a variety of subtle loops to sensory systems, cerebellum, and the basal ganglia. Chapter 8 studies the modular structure and self-organization of visual cortex; we study the role of different thalamocortical oscillatory rhythms in the transition between sleeping and waking; and we model the interaction of multiple cortical regions in vision, saccade control, and cortically guided reaching and grasping. Later chapters then extend our understanding of cerebral cortex by showing how its function can be fully understood only through analysis of its "cooperative computation" with the cerebellum and basal ganglia. Chapter 9 analyzes the role of the cerebellum in both motor control and classical conditioning. finally, Chapter 10 provides an account of the role of the basal ganglia in motor coordination and learning which contrasts its role with that of cerebellum and emphasizes the important role of the dopamine system in its functioning.

In this way, Organization provide a structural analysis of many important brain regions which is integrated with models of a number of the functions these regions serve, both singly and in concert, and of the dynamics of their neural networks. We study a variety of systems involved in sensory analysis (especially for olfaction and vision), rhythm generation, sensory-motor integration (with special attention to visual guidance of eye, arm, and hand movements), and for learning and memory, as well as offering an account of the self-organization of several components of the nervous system.

Part II concludes with Chapter 11, "Prospects for a Neuroscience of Cognition," which both summarizes the progress exhibited in the preceding Chapters, and points the way for the broader use of our methodology in the future development of a Cognitive Neuroscience. Section 11.2, "Multiple Levels, Multiple Methodologies, and the Need for Their Integration", analyzed the themes of this book from a more philosophical, methodological perspective. We started with a discussion of the implications of a non-monolithic approach to the brain involving multiple views and multiple theories. We then examined general issues of brain theory, stressing the transition from the Cartesian reflex paradigm to the self-directed, self-organizing brain, and listing a number of principles of neural organization. The concluding Section, 11.3, "Implications and Outlook For Cognitive Neuroscience", suggests ways in which the ideas developed in this volume may contribute to future work in cognitive neuroscience.

In summarizing the contributions made by Organization, the next three sections of the précis will examine "threads" associated with our three overarching themes of Structure, Function, and Dynamics. As the reader will see, a number of sections of the book are cited under each of these headings, exemplifying our point that these three themes are not separate areas of study, but must be interwoven in the attempt to provide coherent models which integrate and challenge a sufficiently broad range of data. In the same way, the emphasis of Chapters 5 to 10 on one brain region at a time -- Chapter 5, olfactory bulb and olfactory cortex; Chapter 6, hippocampus; Chapter 7, thalamus; Chapter 8, cerebral cortex; Chapter 9, cerebellum; Chapter 10, basal ganglia -- should not hide our essential claim that to understand the brain we must understand how many such regions work together in the integrated production of behavior. In any case, we now summarize the next three sections of this précis , each of which provides a different perspective from which to integrate themes developed throughout the entire volume. In each of these three sections one subsection is expanded in a more detailed way to better illustrate our methodology. (To avoid confusion in what follows, Sec. Px refers to Section x of the Précis , whereas Sec. x simpliciter refers to Section x of Organization.)

Sec. P4 (based on Section 11.1.1) examines the following threads for Structure: (i) phylogeny and ontogeny; (ii) the architectonic basis for analysis of function and dynamics of circuitry; and (iii) the embedding of the region in loops and pathways integrating it with other regions. We note that the transition from structure to function rests crucially (but not only) on "filling in the signs" in the structure as basis for the interplay of excitation and inhibition. We will discuss subsection (ii) more extensively.

Sec. P5 (based on Section 11.1.2) studies the threads of Behavior: (i) the place of schema theory in neuroscience; (ii) high-level constraints on system modeling in general afforded by the study of behavior; and then the use of (iii) the saccadic system and (iv) reaching and grasping to challenge the creation of models of function and learning at the level of both schemas and neural networks. We will present subsection (iv) in a more detailed way.

Sec. P6 (based on Section 11.1.3) reviews the threads woven into our study of Dynamics: (i) the basic dynamic concepts of fixed points, rhythmogenesis and synchronization, and chaos; (ii) self-organization; (iii) plasticity and the modeling of learning; and, finally, dynamics considered at the finer and finer levels of (iv) compartmental modeling, and (v) neurochemistry. In this Section we emphasize subsection (i).

After these three overview sections, Sec. P7 (based on Chapter 6) presents a case study, illustrating the application of our pluralistic strategy to understanding the functional organization and performance of the hippocampus. We show how computational models help to integrate the results of studies at different levels of neural organization (from the subcellular level to the systems level incorporating the hippocampal formation) and of distinct subdisciplines of the neurosciences (from micro- and macroanatomy, and intracellular and EEG-level physiology through animal behavioral and psychological studies).

Finally, Sec. P8 (based on Section 11.3) samples our views on the Prospects for a Neuroscience of Cognition.

In Organization, we have testified to the vitality of a computational neuroscience open to the data of empirical neuroscience by presenting not only our own models but also those of many other researchers. We thus particularly welcome commentaries that present examples of modeling and data collection that carry forward Organization’s program of developing a neuroscience in which the computational and empirical study of neural organization pay attention to all three dimensions of structure, function and dynamics. While we will read with interest commentaries that touch on Szentágothai’s views, we will not in general be able to reply to them at any length in the Response.

P4. Structure

P4.1. Phylogeny and Ontogeny

Sec. 2.1 introduced the idea that the embryology of a structure may make clear crucial relationships that may be obscured in the adult form. We used the segmented part of the neuraxis, i.e., the spinal cord and lower brainstem, to ground discussion of the progressive loss of segmentation in the upper brainstem, including the diencephalon. Moreover, we showed that the embryonic nervous system is able to generate movement before it is able to respond to sensory stimuli, supporting the action-oriented view of brain function stressed in Chapter 3. Turning from ontogeny to phylogeny, we studied the evolution of the mammalian brain, seeing how, as the brain evolves, basic structures become overlaid with more and more complex structures which can both inhibit and coordinate what has evolved before. This evolutionary theme was taken up from a functional point of view in Sec. 3.2, where we showed how brain function can be analyzed in a process of evolutionary refinement of models in which basic systems serve as the substrate for the designed "evolution" of more refined systems - new schemas often arise as "modulators" of existing schemas, rather than as new systems with independent functional roles.

Sec. 7.1 studied an intermediate embryonic stage to illuminate the relationship between the diencephalon and the telencephalic vesicles. The two developmentally different structures are welded together by the progressive emergence of fiber tracts (white matter) formed by the axonal processes of neurons - both ascending and descending. These processes, as in the formation of the spinal cord and the lower brainstem, have the tendency to gather in and to occupy the outer surface areas of the original neural tube. The structural overview of cerebral cortex began with a view of its development (Sec. 8.1), while Sec. 8.2 analyzed the development of two striking examples of modular architectonics in primary visual cortex: ocular dominance columns and orientation columns. Finally, our structural view of the cerebellum was initiated (Sec. 9.1) by a review of its phylogeny and ontogeny, showing how the relative size of different parts of cerebellum may vary from species to species, and tracing the delicate mechanisms of cell growth, migration, and interaction which yield the quasi-crystalline structure of the mature cerebellar cortex.

P4.2. The Architectonic Basis for the Analysis of Function and Dynamics

Sec. 2.2 approached a hierarchy of levels of structural analysis - neurons; networks; integrated system - by presenting data which support Szentágothai’s modular architectonic principle , namely that

Neuronal connectivity in a typical neural center is sufficiently specific to permit disassembly of the whole network into neuronal modules of characteristic internal connectivity; and the larger structure can be reconstituted by repetition of these modules.

Szentágothai analyzed the spinal gray matter and the lower brainstem in these terms, and then showed that the upper diencephalic and telencephalic parts of the brainstem do not retain the quasi-segmental arrangement of the lower neuraxis but that elements of the basic architectural principle of the neuraxis are preserved. Turning from modular structure, we noted the importance in many structures of local circuit neurons, and of the complex synaptic arrangements called glomeruli with their synaptic triads in, e.g., the cerebellar cortex, the olfactory bulb, and some of the anterior thalamic nuclei.

Szentágothai applied the modular architectonics principle to the cerebral cortex, linking observations of anatomical regularities to the observations of Mountcastle on physiological "columns" in somatosensory cortex and of Hubel and Wiesel on visual cortex. However, we (Arbib and Érdi) coined the term "Multiple Models of Modularity" (Sec. 2.3) to stress that the search for a hierarchy of levels of analysis of neurons, networks, and integrated systems is not confined to the "column-like" structures of cerebral sensory cortex. Other important anatomical regularities are the quasi-crystalline structure of the cerebellar cortex and the basic lamellar structure of the hippocampus.

Sec. 5.1 reviewed the layers, cell types, and synaptic organization of the olfactory bulb, and then continued the story into the olfactory cortex, stressing how its laminar structure relates to the afferent fibers from the olfactory bulb, re-excitatory connections which may be the anatomical substrate for olfactory associative memory, commissural fiber systems connecting the two halves of the cortex, and neurochemically varied centrifugal inputs from different brain areas. Sec. 6.1 then provided a structural view of the hippocampus, treating in turn the intrinsic organization - cells and circuits - of the hippocampus, the hippocampal afferents and efferents, both cortical and subcortical , and basic quantitative data on cell numbers, and on the convergence and divergence of connections. Special attention was paid to the "synaptic matrices" revealed in lamellae orthogonal to the long axis of the hippocampus - thus establishing a "basic circuit" for the analysis of hippocampal function.

Sec. 7.1 advanced the theme of modular architectonics by noting that frontally oriented successive discs of the cortex from front to aft have a mutual, although more clearly thalamo-cortical, relationship with close to sagittal discs of the thalamus.

Sec. 8.1 analyzed the synaptic connectivity of cortical neurons, noting the interplay of excitation and inhibition, and the role of neuron chains in cortical function and then returned to the modular architectonics principle with Szentágothai arguing for the modular structural organization of the cortex, but questioning the functional role of such units. Sec. 8.2 offered two striking examples of modular architectonics seen in primary visual cortex: ocular dominance columns and orientation columns.

With this, we turn to a more detailed review of Szentágothai's contributions to Organization, mostly those where he explained his views in historical context, and then we will make some remarks on them in the light of recent results. Since we obviously do not have the competence in neuroanatomy that Szentágothai had, we note once again that while we welcome commentaries on this material, we will not in general be able to respond to them at any length. We start with the concept of neuron chains of the cortex .

The principal question to be considered here concerns the main neuron chain for routing afferent impulse patterns through the particular piece (or region) of cortex under study. Even without exact knowledge of the details of synaptic connectivity, the early diagrams of Ramón y Cajal did convey the essence of the main neuron chains involved. The next crucial step was Lorente de Nó's magnificent abstraction in his Chapter on cortex written for the 1938 edition of J.F. Fulton's Physiology of the Central Nervous System . Although the abstraction went a bit too far in substituting a single bouton for the entire interneuronal connection, two very basic facts were recognized, but only one of them correctly interpreted. One fact recognized by Lorente de Nó was the essentially vertical (up and down) orientation of the neuron chains. The neuronal chains of the cortex are indeed preferentially vertically oriented. Even if some connections for intermediate (1-10 mm) distances are tangential, the final synaptically active parts of the axons are vertically oriented. The other fundamental principle of coupling was called reciprocity and multiplicity of synaptic connections. Although the observation was essentially correct, even if perhaps a bit exaggerated, both Ramón y Cajal and Lorente de Nó had a curious "blind spot" in their field of scientific vision. This is all the more astonishing, because nervous inhibition was postulated already by René Descartes, and was experimentally first observed by the Russian physiologist Sechenov (1863), and thoroughly elaborated upon in the works of C.S. Sherrington (1906). However the word inhibition does not occur according to our knowledge in any of the works of Ramón y Cajal and was explained by Lorente de Nó (1938b) as effected by some kind of Wedensky effect (an outdated concept - ingenious, but basically wrong).

The revival of Golgi studies, especially in the US by scientists like M.A. and A.E. Scheibel, C. Fox, the Lunds and many others) - the Russian neurohistologists (Poljakov, Skolnik-Jaross, Leontovitch) had kept Golgi studies continuously on very high standards - was a prelude to efforts trying to understand neuron chains by a combination of Golgi stains, experimental axon and synapse degeneration, and chronically isolated but vascularized neural tissue block techniques, introduced by Szentágothai. However, this was only a transitory stage towards the more sophisticated approach introduced in the late 1970s and early 1980s by Peter Somogyi (1977; Somogyi et al., 1979, 1981a,b).

Given Szentágothai's attention to the work of Somogyi, we briefly turn here to Somogyi et al.’s (1998) discussion of "Salient features of synaptic organization in the cerebral cortex". Here the attempt is made to define basic cortical circuits on the basis of quantitative studies of the synaptic connectivity of identified cortical neurons and their molecular dissection. By studying the precise location of postsynaptic GABA and glutamate receptors at cortical synapses, Somogyi et al. are able to argue that, due to the exclusion of G protein-coupled receptors from the postsynaptic density, the presence of extrasynaptic receptors and the molecular compartmentalization of the postsynaptic membrane, the synapse should include membrane areas beyond the membrane specialization. Subsequently, they examine five organizational principles:

1. The cerebral cortex consists of: (i) a large population of principal neurons reciprocally connected to the thalamus and to each other via axon collaterals releasing excitatory amino acids, and, (ii) a smaller population of mainly local circuit GABAergic neurons.

2. Differential reciprocal connections are also formed amongst GABAergic neurons.

3. All extrinsic and intracortical glutamatergic pathways terminate on both the principal and the GABAergic neurons, differentially weighted according to the pathway.

4. Synapses of multiple sets of glutamatergic and GABAergic afferents subdivide the surface of cortical neurons and are often co-aligned on the dendritic domain.

5. A unique feature of the cortex is the GABAergic axo-axonic cell, influencing principal cells through GABA A receptors at synapses located exclusively on the axon initial segment.

Having said all this, they find that the basic circuit proves highly adaptable when comparisons are made between cortical areas or layers:

The basic circuit is most obvious in the hippocampus where a relatively homogeneous set of spatially aligned principal cells allows an easy visualization of the organizational rules. Those principles which have been examined in the isocortex proved to be identical or very similar. In the isocortex, the basic circuit, scaled to specific requirements, is repeated in each layer. As multiple sets of output neurons evolved, requiring subtly different needs for their inputs, the basic circuit may be superimposed several times in the same layer. Tangential intralaminar connections in both the hippocampus and isocortex also connect output neurons with similar properties, as best seen in the patchy connections in the isocortex. The additional radial superposition of several laminae of distinct sets of output neurons, each representing and supported by its basic circuit, requires a co-ordination of their activity that is mediated by highly selective interlaminar connections, involving both the GABAergic and the excitatory amino acid releasing neurons. The remarkable specificity in the geometry of cells and the selectivity in placement of neurotransmitter receptors and synapses on their surface, strongly suggest a predominant role for time in the coding of information, but this does not exclude an important role also for the rate of action potential discharge in cortical circuitry." (Somogyi et al. 1998. Copyright 1998 Elsevier Science B.V. All rights reserved.)

It is interesting how Somogyi et al. 1998 emphasized the common organization principles of the basic circuits of the hippocampus and of a single layer of the isocortex. With this we return to Szentágothai’s words: "The analyses possible in the hippocampal formations need further development to lead to an improved insight into the intricacies of the complex network of the many types of inhibitory neurons in the neocortex."

Now we continue with Szentágothai’s discussion of the modular architectonics principle. The concept of a modular architectonics principle arose from two entirely independent sources: (i) the observation by the Scheibels (1958) of certain spatial regularities in the arborization both of dendrites and of axonal ramification in the lower brainstem; and (ii) some kind of "columnar" organizations of the somatosensory cortex (Mountcastle, 1957) and the even more convincing observation by Hubel and Wiesel (1959) of so called "orientation columns" in the visual cortex. While the observations of the Scheibels had their main follow up in our own (Szentágothai) studies in the spinal cord, the observations of Mountcastle and Hubel and Wiesel prompted efforts to adapt new findings in cortical histology to the new insight gained from emergent cortical physiology. Actually, a very first attempt by Szentágothai (1967) anticipated the concept of the "modular architectonics principle" under a different name: "elementary integrative unit". This expression was later abandoned because it was misleading and in fact led to certain misinterpretations (very unfortunately in some otherwise fundamental writings of Sir John Eccles). The first explicit modular neuron connectivity model of the cortex was proposed (Szentágothai 1969) for the somatosensory cortex in a diagram reproduced as Fig. 8.12 in Organization. This model, as well as the earlier 1967 integrative unit model was still under the influence - and mimicking – Szentágothai’s earlier cerebellar cortex model. It took a further 10 years before, still using the earlier "pre-Somogyi" guesswork strategy, obtaining the more realistic model of (Szentágothai 1983).

Though we cannot review the newer data and their interpretations in full detail, we (Arbib and Érdi) will briefly review some recent developments related to modular architectonics. The whole issue seems to be controversial.

Lev and White (1997) studied the organization of pyramidal cell apical dendrites in the primary motor cortex of the mouse. According to their data, apical dendrites of callosal neurons aggregate to form clusters that are composed exclusively of dendrites belonging to this type of projection cell. Thus the cellular composition of cortical modules seems to be much more specific than was thought earlier. Thy also hinted that the sensory cortices may also have such a kind of modular organization.

Based on studies on in the macaque monkey cortex, Britten (1998) reported that columnar organization exists beyond the "early" stages (e.g., area 17) of cortical processing. By analyzing response selectivity, a clustered organization of neurons sharing response properties was found in the medial superior temporal area of extrastriate cortex, which is most likely the highest visual area on this pathway.

Results of recent investigations of the cytochrome oxidase (CO) blobs can by and large be considered as arguments for the existence of columnar organization. Murphy et al. (1998) suggested that in visual cortex of normal and strabismic monkeys the fundamental periodicity of this columnar architecture was formed prenatally, and is not modifiable by experience. Elston at al (1999) found a correspondence between the dendritic fields of layer V pyramidal cells and the CO bands, but they also found many examples where the dendrites crossed the boundary between bands.

Vernon Mountcastle, one of the pioneers of the columnar organization of the neocortex, states (Mountcastle, 1997) that modules may vary in cell type and number, in internal and external connectivity, and in mode of neuronal processing between different large entities, but within any single large entity they have a basic similarity of internal design and operation. A cortical area defined by the rules of classical cytoarchitectonics may belong to different systems. Therefore distributed structures may serve as the anatomical bases of distributed function.

In Organization, we mentioned that Swindale (1990) explicitly criticized the modular architectonic principle and some debate between him and Szentágothai ensued. Swindale (1998) reiterated his interpretation and stated that instead of using the concept of modular organization of the cortex, recent studies reveal "...a more fluid arrangement in which several separate maps are superimposed, with relatively weak geometric linkages and no common modular submit." In fact, Szentágothai considered cortical organization to be the result of the interplay between determinism and chance and never talked about rigid modules.

A few notes complete our tour of the architectonic basis for the analysis of function and dynamics. Sec. 9.2 studied the quasi-crystalline structure and the "space economy" of the cerebellar cortex in some quantitative detail. The only output cells of the cerebellar cortex are Purkinje cells, and these inhibit nuclear cells. Thus we stressed the integration of the circuitry of the cerebellar cortex into a "microcomplex" which unites a microzone of cortex with the region of nucleus to which it projects, and with which it shares afferents - thus establishing a "basic circuit" for the analysis of cerebellar function. Sec. 10.1 examined the structure of the striatum, the region of the basal ganglia which receives most of its input pathways: by looking at the two pathways, direct and indirect, whereby it acts upon the output pathways of the basal ganglia; and by looking at the division of the striatum into "patches" embedded in a "matrix".

P4.3. The Embedding of Regions in Loops and Pathways

Sec. 2.3 emphasized that many regions of the brain are best thought of as embedded within even larger systems integrated by loops traversing many brain regions. Consider, for example, the links of the cerebellar system - "upstream" with the cerebral cortex and "downstream" with the spinal cord - which are closed in the cerebellar nuclei to which the output cells of cerebellar cortex project. The fact that the output of cerebellar cortex is purely inhibitory ties into the theme that the passage from structure to function is often based on understanding the patterns of interplay of excitation and inhibition. In this case, the inhibition from cerebellar cortex serves to modulate the activity in the cerebellar nuclei, which serve in turn to tune and coordinate motor pattern generators located elsewhere in the nervous system.

The role of the thalamus as the chief relay for sensory input to cerebral cortex is only a small fraction of the crucial role of the thalamus in all kinds of pathways. We thus devoted Sec. 7.3 to thalamo-cortical loops and cooperative computation, reviewing the descending control of sensory systems in the lateral geniculate nucleus of mammals, as well as thalamocortical oscillations. We then briefly reviewed the thalamo-cortical loops involving the basal ganglia and cerebellum. Sec. 9.3 then stressed that the microcomplexes mentioned above integrate cerebellar cortex into a cerebellar system which modulates motor pattern generators (MPGs). Sec. 10.1 showed the basal ganglia to be embedded in four disjoint loops of the form "cortex -> striatum -> SNr (substantia nigra pars reticulata) -> thalamus -> prefrontal cortex": the oculomotor circuit, the motor circuit, the "cognitive" (dorsolateral prefrontal) circuit, and the limbic (lateral orbitofrontal) circuit, providing our concluding example of the embedding of regions in loops and pathways.

P5. Function

P5.1. The Place of Schema Theory in Neuroscience

Chapter 3 presented schema theory as a framework for the rigorous analysis of behavior which requires no prior commitment to hypotheses on the localization of each schema (unit of functional analysis), but which can be linked to a structural analysis as and when this becomes appropriate. Sec. 3.1 introduced an approach to schema theory which emphasizes action-oriented perception, with the paradigm of the action-perception cycle replacing the stimulus-response paradigm. However, a number of later sections have made clear that schema theory and the action-perception cycle - and our approach to functional neuroscience in general - are not limited to those forms of, e.g., sensorimotor coordination for which extensive neural data are available, and which we treated at length in Part II. These sections include Sec. 3.1.5. on "Visual Scene Interpretation", Sec. 6.5 on "Hippocampal Function and Human Memory", and Sec. 8.6."From Action-Oriented Perception to Cognition", and the discussion continued in Secs. 11.3.4. "Schema Theory and the Construction of Reality" and 11.3.5, "Language".

Sec. 3.1 explored the constraints imposed by linking schema theory to functional neuroscience, and provided a quasi-formal introduction to perceptual and motor schemas, coordinated control programs (illustrated with an introduction to the visual control of reaching and grasping), cooperative computation, and schema assemblages (the basis for a schema-based model of visual perception which provided a perspective on short-term and long-term memory). A simple account of approach and avoidance behavior in frogs illustrated the use of perceptual and motor schemas, showing how they may be linked; analysis of lesion data illustrated the issues involved in making a schema-based account of a function into a brain model.

P5.2. Behavioral Constraints

Sec. 3.2 presented Rana computatrix , a set of models of visuomotor coordination in frog and toad, studying approach, avoidance and detour behavior to show how perception may demand the mutual refinement of one perceptual schema by another, how multiple motor schemas may act together to yield complex motor behaviors, and how brain function can be analyzed in a process of evolutionary refinement. We then studied neural mechanisms of avoidance behavior to provide our first example of how neural modeling can be used to replace schemas with neural networks of equivalent functionality.

Secs. 3.3 and 3.4 introduced schemas for looking, reaching, and grasping to demonstrate that much is to be learned at the level of schema analysis prior to, or in concert with, the analysis of neural circuitry. These schemas were shown, in Chapters 8 and 10, to be distributed across cerebral cortex and basal ganglia, while Chapter 9 showed the role of cerebellum in their adaptation and coordination.

Sec. 6.4 combined a functional view of the hippocampus - its role in the cognitive maps underlying navigation and spatial behavior in rats - with a dynamic view of how synaptic plasticity may enable hippocampal cells to learn to encode different "places" in a cognitive map. We offered a general framework for the study of spatial representation and cognitive maps in rats, including the general idea of World Graphs as cognitive maps for motivated behavior, and then reviewed the neurophysiology of spatial representation, with special emphasis on the "place cells" of the regions CA3 and CA1 of hippocampus. We offered two contrasting systems views of the role of the hippocampus in navigation, in each case emphasizing that the representation of current place in CA3 and CA1 is insufficient for a cognitive map which underlies navigation. Sec. 6.5 viewed the role of hippocampal function in human memory, introducing the crucial dichotomies of procedural vs. declarative memory and of skill vs. episodic learning. The data suggest that the hippocampus is involved in declarative rather than procedural memory and in episodic rather than skill learning. We closed the Chapter by discussing the very much open question: "Is there a commonality of mechanism between the two main functions attributed to the hippocampus: cognitive mapping in rats and declarative memory in humans?" (Since learning is a crucial aspect of adaptive behavior, a number of the issues discussed here under the heading of "Function - Behavioral Constraints" overlap those discussed below under the heading "Dynamics - Plasticity: Modeling Learning".)

Sec. 8.2 offered a very brief look at how primary visual cortex provides input to a variety of visual processes, introducing psychophysical and neurophysiological data on spatial visual perception, and studying the role of long-range horizontal connections in the integration of information. We had little more to say about low-level visual processing, but Sec. 8.4 did discuss cortical mechanisms for using vision in the control of movement. That section focuses on frontal-parietal interactions in cortex, but many other brain regions are involved in the integration of vision with action. Sec. 8.5 turned to the theme of learning of coordinated behaviors, providing both a schema-level analysis of motor set and the neuralization of coordinated control programs, and a specific neural network model of visual-motor conditional learning. This theme, foreshadowed in the study of rat spatial learning (Chapter 6), was further developed in our study of cerebellum (Chapter 9) and basal ganglia (Chapter 10). Sec. 8.6 then charted basic processes underlying cognitive functions from a high-level schema-theoretic viewpoint.

Sec. 9.3 focused on the issue of how skills are acquired, viewing the cerebellum as a learning machine, stressing the idea that cerebellar nuclei modulate motor pattern generators (MPGs) while the cerebellar cortex learns how best to modulate the cerebellar nuclei: modulating the modulator. We first studied the role of the cerebellum in the vestibulo-ocular reflex (VOR) where data strongly support the notion of a functional role of an adaptive microcomplex in modulating the gain of eye movements that compensate for head movements, and in the classical conditioning of the rabbit eye-blink response.

Sec. 9.4 then presented models of how the cerebellum adapts the metrics of movement to changing circumstances. We showed how the detailed circuitry of the cerebellar cortex and of various nuclei with which it interacts could modulate activity in MPG-related loops on a short-term basis, i.e., one appropriate to the current circumstances as in adjusting to step height in climbing a flight of stairs. The remaining subsections reviewed approaches to modeling adaptation of motor control where the adaptation persists on a long-term basis, involving synaptic plasticity: we presented feedback-error learning whereby the cerebellum could "take over" motor control from other parts of the brain, but argued that the cerebellum "works" by modulating and coordinating multiple Motor Pattern Generators (MPGs), rather than by replacing them.

In addition to behavior in normal subjects, we can learn much about neural function and dynamics by studying their fate in subjects with a variety of diseases. Sec. 9.3 noted clinical data on the role of the cerebellum in skilled movements as a basis for our study of how these skills are acquired, viewing the cerebellum as a learning machine. Sec. 10.2 focuses on diseases of the basal ganglia, showing that the distinct movement disorders seen in Huntington's disease (hyperkinetic and hypotonic) and Parkinson's disease (hypokinetic and hypertonic) are associated with decreased basal ganglia output in Huntington's disease and a marked increase in Parkinson's disease.

P5.3. The Saccadic System

Sec. 3.3 introduced schemas for controlling the rapid eye movements called saccades, and focused on the homology between the tectum in frog and toad and the superior colliculus in primates - the whole body movement of the frog towards its prey corresponding to the orienting of gaze towards a visual target in the monkey. It showed how schemas for Working Memory and for Dynamic Remapping may extend the monkey's saccadic repertoire to include saccades to remembered targets or to two targets in succession.

Later sections then showed how circuitry in various regions of the brain may contribute to these and other schemas. Sec. 8.4 presented a detailed model of cortico-thalamic systems for saccade control, with particular attention to the roles of posterior parietal cortex, frontal eye fields and thalamus in providing mechanisms for dynamic remapping and working memory. Sec. 9.4 modeled the cerebellar role in saccade adaptation, refining the cortico-thalamic model by adjusting the metrics of saccades, a feat of learning that is impossible in animals or humans lacking certain portions of the cerebellum. Sec. 10.4 then completed our model of saccade control by focusing on the "oculomotor loop" of the basal ganglia, presenting a model of the role of the entire loop in generating saccades, and modeling the interactions between the basal ganglia and the "working memory" systems of prefrontal cortex. Interruption of SNr (substantia nigra pars reticulata) inhibition allows reciprocal connections between frontal eye fields and thalamus to generate a spatial "memory" cycle or loop. Once a saccade has been made to a remembered target, the memory trace must be erased to prevent generation of further saccades of equal magnitude and direction. We posit that the activity of this spatial working memory could be regulated by the inhibitory topographic projection from SNr to thalamus.

Sec. 10.6 suggested that if the frontal eye field and direct visual input do not yield the encoding of a unique target in the deep layers of SC (superior colliculus), then a winner-take-all (WTA) mechanism will "choose" one of the targets. We then argued that experience based on inferotemporal or prefrontal information may provide contextual, learned information to bias activity in the basal ganglia to "tip the balance" to one "winner" or another, presenting models of visual-motor conditioning, including spatial generalization and sequential behavior based on the strong hypothesis that this learning is mediated by cortico-striatal plasticity.

P5.4. Reaching and Grasping

In studying the role of perception in mediating behavior, we stress that there is in general no complete and objective "percept" of an object, but rather a set of partial characterizations (including parameters that we may not be able to symbolically represent in any explicit fashion) related to the current set of goals and motivations of the observer - which may keep unfolding as interaction with, or contemplation of, the object continues. Sec. 3.4 illustrated this using the schemas involved in reaching and grasping. We presented the concepts of virtual fingers and opposition space to offer a precise but compact description of the degrees of freedom involved in a number of grasping movements, and then turned to a series of experiments which motivated the design of a new coordinated control program that explicitly involves a coordinating schema as well as perceptual and motor schemas.

Sec. 8.4 turned to cortical systems for reaching and grasping, and we shall discuss this work in some detail. The work of Ungerleider and Mishkin (1982; Mishkin, Ungerleider, & Macko, 1983) distinguished two visual systems in extrastriate visual processing: the ventral system, V1 -> V2 ->V4->IT (inferotemporal cortex), is characterized as the cortical what (pattern recognition) system; and the dorsal system, extending from V1 to PP (posterior parietal cortex) is characterized as the cortical where (object location) system. Goodale & Milner (1992) reviewed a variety of data including those on the ability of a patient with a ventral lesion to carry out a variety of object manipulations even though unable to verbally report on the object parameters used to guide these actions. They concluded that the dorsal system mediates the required sensorimotor transformations for visually guided actions directed at such objects and so extended the Mishkin-Ungerleider dichotomy to view the dorsal system as the how system since location (where) is only one of many properties needed to determine how to interact with an object.

P5.4.1. The FARS Model of Parietal-Premotor Interactions in Grasping: We have developed a detailed model, the FARS (Fagg-Arbib-Rizzolatti-Sakata) model, based on the interactions between the AIP (anterior intraparietal sulcus) area of PP, and the F5 area of premotor cortex in monkeys trained to grasp objects. About half the neurons related to hand movements in AIP fired almost exclusively during one type of grip, with precision grip being the most represented grip type (Taira et al. 1990; Sakata et al. 1992). Some cells demonstrate specificity toward the size of the object to be grasped; and some cells demonstrated independence from the size of the object. A few cells show modulation based upon the object's position and/or orientation in space. The visual responses of these cells thus provide a distributed code for affordances for grasping, i.e., the various ways in which an object may be grasped (as distinct from recognizing what the object is). Most neurons in AIP also show phasic activity related to the motor behavior. Five identifiable phases occur in the paradigm used by Sakata to study these cells: set ( key phase ), preshape, enclose, hold ( object phase), and ungrasp. Cells participate in varying degrees during different phases of the movement, but are usually most highly active during the preshape and enclose phases of movement. Very importantly, once an AIP cell becomes active, it typically remains active until the object is released.

The main anatomical connections of F5 are with AIP and the hand field of the precentral motor area (Muakkassa & Strick 1979; Matelli et al. 1985). Rizzolatti et al. (1988) described various classes of F5 neurons which discharge during specific hand movements (e.g., grasping, holding, tearing, manipulating). The largest class is related to grasping. The temporal relations between neuron discharge and grasping movements vary among neurons.

We now outline the FARS model, implemented in terms of simplified but biologically plausible neural networks (Fagg and Arbib 1998). Given visual input from an object, the model AIP computes its affordances. The corresponding set of grasps is passed to F5. As a function of task or other information, F5 selects one of the specified grasps, and is responsible for unfolding the grasp in time. F5 activity is broadcast back to AIP strengthening the affordance that corresponds to the selected grasp. Motor responses in AIP are explained as corollary discharges from F5, and AIP provides an active memory for the grasp which is continuously updated. This is similar to the dynamic remapping mechanism in our study of saccades ( Organization Sec. 8.4.2), in which motor afference updated a map of targets of potential eye movements.

The location of target objects is passed to F4, which represents the arm goal position. Since grasp programming affects arm movements, the model modulates F4 with information from AIP specific to the affordance/grasp pair selected by the AIP/F5 system. A neighboring region, the posterior intraparietal area (PIP) codes object-centered information (Sakata, personal communication) concerning different shapes presented to the monkey. In the model, PIP codes the shape and size of the object to be grasped. An affordance derived from PIP maps an object configuration to one possible grasp for that object. Castiello et al. (1991) studied impaired grasping in a patient (AT) with a lesion impairing the pathway V1->PP, and found evidence for a mapping from object identity to affordances that is effective whenever the nature of the object such a mapping. The model thus includes a corresponding path PIP -> IT -> AIP.

The FARS model analyzes the interaction between AIP and F5 populations during execution of the Sakata paradigm. AIP units include visual-related cell that recognizes objects that require a specific grasp and motor-related cells which are active for specific grasps. Each F5 unit fires during a different phase of the program. At each program phase, the state is reported back to the AIP motor-type population. The full model also includes the role of SII in creating and monitoring haptic expectations, the role of dorsal premotor cortex (F2) in the association of arbitrary stimuli with motor program preparation, and the role of area 46 as a working memory in tasks requiring information to be held during a delay period. However, these details (and the presentation of simulation results) were beyond the scope of Organization (see Fagg and Arbib 1998 for details). We do stress, however, that the circuitry controlling F5 programs in the model is not intrinsic to F5: the effective connections between program states are not coded within F5 but are managed by the combined action of pre-SMA (F6) and the basal ganglia (BG).

One simulation study showed how the model performs when a delayed instruction stimulus is used to inform a subject how to grasp an object. The model is presented with a single object (a small cylinder), and asked to perform one of three tasks. The 3 different tasks are:

1. Grasp the cylinder using a precision pinch;

2. Grasp the cylinder using a side opposition; and

3. As a function of an instruction stimulus (e.g., the color of a light), grasp the cylinder using either a precision pinch or a side opposition.

(We speak rather loosely here. The model is not a robot. It transforms visual codes for objects to neural codes for movements via neural network models of diverse brain regions.) In the model, area F2 (dorsal premotor cortex) has a high level of activity in the conditional task as this region is only involved when the model must map an arbitrary stimulus to a motor program (in this case, a grasp); the region does not receive IS (instructional stimulus) inputs in the non-conditional task. F5 receives inputs from F2, causing an increase in the region's activity level which is passed on through excitatory connections to AIP.

P5.4.2. Synthetic PET imaging : In further work, we sought to understand how our study of the monkey could be related to the results of human brain imaging. In human, PET and fMRI techniques allow us to achieve a global view of the systems involved in performing a task, but at the expense of a very coarse spatial and temporal resolution. In monkey, on the other hand, we are able to examine individual cells and resolve single spikes, but have tremendous difficulty in examining entire circuits. It is thus important to develop techniques that allow experimental results at both levels to be brought together as we attempt to understand the different systems. Arbib, Bischoff, Fagg, & Grafton (1995) proposed Synthetic PET imaging as a way to draw conclusions in one domain from experimental results in the other. The synaptic activity of a region A during a task is computed as the measure of instantaneous synaptic activity in region A integrated over the time required to perform the task (which might involve multiple trials). The simulated synaptic activity of a region can then be compared over several conditions.

With this background, we summarize one prediction made by applying Synthetic PET to the FARS model. The conditional task is to grasp a cylinder using either a precision pinch or a side opposition, the choice being determined by an instruction stimulus (the color of a light). We tabulated regions in the model that demonstrate a change in synaptic activity in the conditional task above and beyond those involved when the subject knows a priori which of the two grasps to perform. The most significant change predicted by the model is the level of activity exhibited by area F2 (dorsal premotor cortex). Its high level of activity in the conditional task is due to the fact that this region is only involved when the model must map an arbitrary stimulus to a motor program (in this case, a grasp). In the non-conditional task, the region does not receive IS inputs, and thus its synaptic activity is dominated by the general background activity in the region. The additional IS inputs in the conditional task have a second-order effect on the network, as reflected in small changes in activity in F5, BG, and AIP. The increased synaptic activity in F5 is due to the additional inputs from F2 (into the supporting inputs of some columns in F5). These inputs also cause an increase in the region's activity level , which is passed on through excitatory connections to AIP.

The above synthetic PET experiments raised some important questions about how instruction stimuli are mapped to arbitrary motor programs, and about the relative representation of different grasps. These predictions were tested in a human PET experiment (Grafton, Fagg, and Arbib, 1998).

The model predicts that the conditional task should yield much higher activation in F2 (dorsal premotor cortex), some activation of F5, and a slight activation of AIP. The human experiment confirmed the F2 result, but failed to confirm the predictions for F5 and AIP. In fact, in humans there is an activation of the inferior parietal cortex (AIP), but no significant activation of ventral premotor cortex. The model involves reciprocal connections between regions F5 and AIP, and a projection from F2 to F5 - but the strength of the projection from F2 to F5 is essentially a free parameter of the model: there is a wide range of values over which the model will correctly perform the conditional and non-conditional tasks. The implication is that, by tuning this parameter, we can control this projection's contribution to the synaptic activity measure in F5. However, the original FARS model is such that difference in AIP synaptic activity from the non-conditional to the conditional task will always be less than the difference observed in F5. One possibility for repairing this problem in the model is to reroute the F2 information so that it enters the grasp decision circuitry through AIP, rather than F5.

The low-level details of the FARS model were derived primarily from neurophysiological results obtained in monkey. The Synthetic PET approach extracts measures of regional synaptic activity as the model performs a variety of tasks. These measures are then compared to rCBF (regional cerebral blood flow) observed during human PET experiments as the subjects perform tasks similar to those simulated in the model. In some cases, the human results provide confirmation of the model behavior. In other cases, where there is a mismatch between model prediction and human results, it is possible (as we have shown) to use these negative results to further refine and constrain the model and, on this basis, design new experiments for both primate neurophysiology and human brain imaging. Our point here is not to highlight (or hide!) the flaws in the present FARS model, but rather to suggest that it provides a useful platform for further modeling, and that Synthetic PET provides a technique (itself open to fruitful modification) to ensure that the future modeling is responsive, and contributory, to future developments in both monkey neurophysiology and human neurology.

P5.4.3. Back to the "guided tour": Sec. 9.4 then modeled the role of cerebellum in adapting a particular class of arm movements - the adaptation of dart throwing to wearing prisms. Sec. 10.5 examined the still somewhat discordant views on the roles of the basal ganglia in motor coordination and learning, emphasizing the view that the basal ganglia receives rich contextual information and then "releases" components of motor programs through disinhibition. We suggested that basal ganglia neuronal responses are task-dependent, with the current responsiveness of a neuron possibly determined by the state of behavioral experience or learning, and temporarily maintained until the relevant memory is formed in premotor cortices. We emphasized "channels" involved in working memory and in initiation of movement. The basal ganglia were seen, in part, as controlling a kind of "working memory" for coordinated control programs of motor schemas.

Roles for basal ganglia inhibition may include focusing (suppression of inappropriate movements), sequencing (suppression of forthcoming movement during preparation), and "simulation" in the sense of suppressing motor areas to allow activity in association cortex to be "disconnected", yielding covert "mental simulation" without immediate overt movement. We sought to distinguish the roles of basal ganglia and cerebellum, suggesting that basal ganglia are involved in explicitly combining the "pieces" that make up a skilled behavior, while the cerebellum serves to turn a procedure into a skill, adjusting parameters to adapt and coordinate components of the movement to yield a seamless whole.

P6. Dynamics

P6.1. Fixed points, Rhythmogenesis and Synchronization, and Chaos

Chapter 4 stressed that neural systems can be studied at different levels, such as the molecular, membrane, cellular, synaptic, network, and system levels. Moreover, we noted two main neurodynamical problems: study of the dynamics of activity spreading through a network with fixed wiring; and the study of the dynamics of the connectivity of networks with modifiable synapses - both in normal ontogenetic development, and in learning as a network is tuned by experience. We introduced the key dynamical concept of an attractor, a pattern of activity which "captures" nearby states of an autonomous system. An attractor may be an equilibrium point, a limit cycle (oscillation), or a strange attractor (chaotic behavior). We also looked at the structure-function problem: for what overall patterns of connectivity will a network exhibit a particular temporal pattern of dynamic behavior? The results given were suggestive rather than directly applicable to biologically realistic models of neural networks. Sec. 4.5 introduced Hopfield networks to show work on neural networks motivated by statistical mechanics, including ideas of "energy", "temperature", and the statistical distribution of patterns in relation to an attractor-based model of pattern recognition - and then gave a critique of "computation with attractors".

Sec. 4.2 introduced the topic of oscillatory behavior in neural systems: single cell oscillations resulting from the interplay of a few currents; and central pattern generators (CPGs) in which network of neurons can produce rhythmic behavior in the absence of sensory input. Bifurcation analyses were used to show a transition from equilibrium point to small amplitude oscillation, or from oscillation to chaos, as some control parameter passes through a critical value. We studied phase lags in chains of oscillators (mimicking data on the spinal cord of the lamprey), the importance of long-range coupling in the synchronization of more fully coupled networks (as in models of cortical structures), and bifurcation analysis of gait transitions in locomotion.

Sec. 4.3 studied chaotic behavior in the nervous system. Chaotic systems are characterized by sensitivity to initial conditions. We showed that the structural conditions of chaos occur at different hierarchical levels of neural organization. Neurochemical synaptic transmission is often characterized as a random process, but the "dripping faucet" model may be adapted to explain this apparent randomness as a case of deterministic chaos. We found that global cortical dynamics, as seen at the global level in the electroencephalogram (EEG), may also exhibit chaotic behavior. We discussed "dynamical diseases", introducing the diagnosis and also the control of chaos associated with normal and pathological brain functions. We also discussed the possible - controversial but intriguing - functional roles of chaos in normal brain activity, including perception and memory formation.

Sec. 5.2 focused on the dynamics of activity in the olfactory system, which we shall reiterate here in a fairly detailed way. The demonstration of oscillatory activity in olfactory systems (Adrian 1942, 1950) was one of the first experiments to illustrate stimulus-induced activity in the mammalian central nervous system. Phenomenologically, there are two main types of rhythmic activity in the olfactory system: slow and fast oscillations. Slow oscillations around 5Hz may be imposed on the olfactory bulb by the respiratory nuclei (Freeman 1991), or may be induced in the olfactory cortex by cholinergic antagonists (Biedenbech 1966; for further details see Sec. 5.2.3.). The "respiratory wave" in the olfactory bulb is generated by the granule cells in response to input from the receptors through the periglomerular and mitral and tufted (M/T) cells. It can be detected in the mucosa by volume conduction. The slow background activity is phase locked with the respiratory wave and it is identified with the sniff cycle. A sniff cycle is composed of an inhalation and exhalation stage, and its duration is 200-500 msec for rabbits. A slow potential evoked by odorants (Ottoson 1959) appeared also in the electro-olfactogram, a receptor potential recorded in the nasal mucosa, which can spread through the brain by volume conduction.

Electroencephalogram (EEG) patterns also show fast oscillations with frequency 35-90Hz in different parts of the olfactory bulb (Freeman 1978, Freeman & Schneider 1982). (The terminology here is: slow ~1 Hz, intermediate ~10 Hz, and fast ~40 Hz.) It was suggested by pioneering work of Freeman (1975) that the spatiotemporal activity patterns of the olfactory bulb can be interpreted within the framework of dynamical system theory. Later it was suggested (e.g., Skarda & Freeman 1987) that sensory information was encoded in spatiotemporal periodic and chaotic patterns. Generally, field potentials do not provide sufficient information about the underlying neural mechanisms. Still, one important message of Freeman's experimental and theoretical work is that rhythmic-arrhythmic bulbar activity is the result of the interactions between excitatory (mitral) and inhibitory (granule) cell populations. The situation, however, might be more complex. Odor-induced (mitral cell) activity is under GABAergic control (Duchamp-Viret et al. 1993). Even in the case of blockade of the GABA A mediated inhibitory effect of the granule cells, oscillatory bulbar activity may occur as a consequence of recurrent excitatory connections.

Both network-level and detailed single neuron modeling techniques have been used to probe the structural bases of the generation of different rhythms and spatiotemporal patterns. First, a number of network studies are based on single-compartment models in which single cell activity is characterized by an internal state defined by the intracellular membrane potential, and by an output expressed as a firing frequency. Modeling illustrates how the interactions among excitatory mitral cell and inhibitory granule cell populations may generate oscillatory and more complex temporal patterns (Li &Hopfield 1989, Érdi et al. 1993, Aradi et al. 1995). More specifically, given the model anatomical structure and the related set of ordinary differential equations, the first question to be answered is what qualitative dynamical behavior emerges in the parameter space. The nature of attractors, the parameter windows belonging to them, and the bifurcation sequences are determined through systematic (numerical) studies. These results showed that chaos can only be found when there is sufficient lateral excitation. It has also been observed that all neurons oscillate in phase in each periodic region (and also during damping oscillation to a stable focus) and no wave phenomena have been detected in the parameter range studied here.

In other studies, mitral and granule cells were the subject of detailed single neuron modeling (Bhalla and Bower 1993, Aradi and Érdi 1996). Four types of problems of signal generation and propagation have been studied:

(i) The effects of the individual currents and their role in the generation and suppression of action potentials, and in the control of firing frequencies (intracompartmental studies).

(ii) Signal propagation through the compartments of both the mitral and granule cells have been simulated. The effects of both orthodromic and antidromic stimulation have been demonstrated.

(iii) The excitatory-inhibitory coupling between the mitral and granule cells through dendro-dendritic synapses and the effects of the (partial) blockade of the GABAergic inhibition have been shown.

(iv) Dynamic behavior of a skeleton network of the bulbar circuitry taking into account even the periglomerular cells has been studied.

What kind of explanations can be obtained by using such techniques? One example of the dynamic effect of the self-excitation between mitral cells was that GABA antagonists produce prolonged depolarization in the mitral cells (Nowycky et al1981, Nicoll and Jahr 1982), and the reentrant excitation in the mitral layer may be associated to a particular mode of bulbar rhythmogenesis. In the case of two mitral cells connected by mutual excitatory couplings, synchronized burst activity may appear (see Fig. 5.10B in Organization). These simulation results are in accordance with the physiological findings and the suggestion that the blockage of GABAergic inhibition controls the odor-induced activity (Duchamp-Viret 1993). According to the simulation experiments, both the mitral-granule feedback loop and the self-excitation in the mitral layer may provide the anatomical substrate of bulbar rhythmogenesis.

We also modeled rhythmic activity in the olfactory cortex. Similarly, Sec. 6.2 gave a dynamical analysis of electrical activity patterns in the hippocampus, addressing data on the normal electrical activity patterns known as theta rhythms and sharp waves, and the abnormal electrical activity exhibited in epileptic seizures. We also compared the hippocampus with the olfactory system, with special attention to the neural mechanisms of rhythm generation and synchronization.

Thalamus and cortex are highly interconnected by reciprocal projections, giving rise to characteristic dynamic patterns. High frequency rhythms are associated with the waking state, while low-frequency rhythms are associated with sleeping. Sec. 8.3 analyzed the balance between oscillations intrinsic to single neurons and network properties in the generation of thalamocortical oscillations. We analyzed the intrinsic electrophysiological properties of thalamic neurons, thalamocortical neurons, and reticular thalamic neurons, and then studied the dynamics of spindle oscillations, and of delta and slow sleep oscillations. We analyzed the role of brain stem control and cellular mechanisms in thalamocortical activation, and closed by using a number of models to explore the role of single cell dynamics versus emergent network properties.

P6.2. Self-Organization: Modeling Development

Both ontogenetic development of neural structures and their plastic behavior are often considered as dynamic processes in the state space of synaptic connections. The "self-organization" of the nervous system is in general a broader process, including addition as well as removal of synapses, and the modification of synaptic strengths. Self-organizing mechanisms are related to normal ontogenetic development (this subsection) and learning (see the next subsection).

Sec. 4.4 focused on retino-tectal connections, discussing the following issues: specificity versus plasticity; genetically prespecified versus environmentally controlled wiring; marker theories versus activity-dependent mechanisms; decrease of synaptic strength by normalization rule only or by selective mechanisms; deterministic versus stochastic models; sets of discrete nerve cells versus continuous neural fields; and positional information.

Sec. 8.2 showed that modular architectonics may be seen as a pattern of organization resulting from the dynamics of self-organization rather than being completely laid down in the genome. In particular, it provided models of the development of two examples of modular architectonics in primary visual cortex: ocular dominance columns and orientation columns. Sec. 10.3 discussed self-reorganization of the striatum, both the self-organizing character of the pattern formation for striatal compartments, and the relationship between the modular remapping architecture, the tonic firing of certain striatal neurons and their role in coordinated motor behavior.

P6.3. Plasticity: Modeling Learning

Whatever the model of the individual neuron, neural tissues may be modeled as networks of intricately connected neurons in which strengths w ij of the synaptic connections may themselves be described by differential (or difference) equations. These "learning rules" were introduced in Sec. 4.5. These included Hebbian learning and its variations, which include means to avoid saturation of synaptic strengths, ways to accommodate various time delays, differential learning mechanisms, as well as "anti-Hebbian" rules to describe features of dissociations of patterns. We studied synaptic matrix models of associative memory, but also saw how invariant pattern recognition may be modeled using the dynamic link architecture in which Hebbian plasticity is invoked on a fast time scale.

Sec. 5.3 reviewed learning and plasticity in the olfactory bulb and olfactory cortex. Building on the relation of different attractor regions to different lateral connection strengths, we showed how synaptic modification can induce transitions between these regions. Another study of the olfactory bulb models associative memory, and shows that incomplete input patterns due to lower odor concentrations can also be identified as proper stimuli if a suitable learning rule is used to modify the lateral connections between mitral cells. We presented two "scenarios" for learning and memory in the olfactory cortex. One was based on the observation that the "sniffing rhythm" of 5Hz may be optimal for inducing long-term potentiation (LTP) in olfactory cortex, and described a hierarchical clustering of input stimuli. The other was based on the argument that the mechanism of object recognition in the olfactory cortex is close to those offered by abstract associative memory models, emphasizing that the incoming (bulbar) information has a complex, distributed representation while the intrinsic excitatory connections between pyramidal cells are spatially extensive, overlapping, and modifiable.

Sec. 6.4 used the role of the hippocampus in the cognitive maps underlying navigation and spatial behavior in rats to ground a dynamic view of synaptic plasticity, showing how Hebbian-like plasticity may enable hippocampal cells to learn to encode different "places" in a cognitive map. We reviewed various neural network models of place cell training, allocentric location, and navigation, one of which pays special attention to data relating place cell activity to the theta rhythm, i.e., relating the dynamics of rhythmogenesis to the synaptic dynamics of learning.

Sec. 8.5 studied the learning of coordinated behaviors, providing both a schema-level analysis of motor set and the neuralization of coordinated control programs, and a specific neural network model of visual-motor conditional learning.

Sec. 9.3 focused on models of the cerebellum as a machine for learning motor skills, starting with the Marr-Albus model which views the Purkinje cell (the output cell of cerebellar cortex) as a perceptron, noting that data on long-term depression (LTD) support the Albus version of the model, namely that "coincidence" of climbing fiber and parallel fiber activity on a Purkinje cell depresses the efficacy of the synapses of parallel fibers active during the conjunction. We stressed the idea that cerebellar nuclei modulate motor pattern generators (MPGs) while the cerebellar cortex learns how best to modulate the cerebellar nuclei: modulating the modulator. Sec. 9.4 reviewed approaches to modeling adaptation of motor control where the adaptation persists on a long-term basis, involving synaptic plasticity (with the emphasis remaining on LTD of parallel fiber Purkinje cell synapses).

Sec. 10.6 argued that experience based on inferotemporal or prefrontal information may provide contextual, learned information to bias activity in the basal ganglia to "tip the balance" to one course of action or another. We presented models of visual-motor conditioning, including spatial generalization and sequential behavior based on the strong hypothesis that this learning is mediated by cortico-striatal plasticity which mediates a form of reinforcement learning in which dopamine released by the SNc acts as the reinforcement signal to toggle between Hebbian and anti-Hebbian learning.

P6.4. Compartmental Modeling

Sec. 4.2 introduced some of the specific formalisms used to treat neurons and neural networks as dynamic systems. The framework for the detailed treatment of the dynamics of the membrane potential of a patch of neuron is provided by the neuronal cable equation and the Hodgkin-Huxley equation, and its relatives.

A whole neuron may either be modeled by a multi-compartment model with compartments chosen to take into account the location of the entering synaptic currents or the geometry of dendritic branching, say, or as a single-compartment model characterized by a single membrane potential. The leaky integrator neuron is a popular model for the single-compartment case.

Sec. 8.4.1 provided the formalism for large-scale models of the nervous system used in many of the models which are based on simple (single-compartment) models of neurons. We now review cases where compartmental modeling has already yielded additional insights. Both periodic and chaotic temporal patterns can be generated at the single neuron level (Sec. 4.3). Basic phenomena can be modeled with membrane equations involving two functionally distinct currents, the slow and fast currents, in which a series of complex patterned activities (simple slow oscillation, bursting, bursting-chaos, beating-chaos and beating) can be generated by changing the time constant of inactivation of the slow current.

In Sec. 5.2.2, in addition to modeling a network built from "integrate-and-fire" elements, multi-compartmental models were given for the mitral and granule cells: six and four compartments were taken into account, respectively. This demonstrated specific effects of the individual ionic conductances on the overall performance of the compartment, signal propagation through the compartments, and synchronization in small networks. Sec. 5.2.3 presented the Wilson-Bower-Hasselmo model of temporal patterns in the piriform cortex which uses a five-compartment model for each pyramidal cell and explicit delays for transmission and axonal activity to clarify the assumptions leading to near 40Hz cortical oscillations. By contrast, the Liljenstrom-Hasselmo model is designed to simulate modulatory cholinergic effects. Their network is built from relatively simple units whose output depends on a factor Q designed to represent the level of acetylcholine. Depending on the values of Q, the system may exhibit convergence to a fixed point, limit cycle oscillation, or (at least transient) chaotic behavior. Moreover, the strengths of the synaptic connections can also drastically influence the dynamic behavior.

Sec. 6.2 presented multi-compartmental neuron models of pyramidal cells and interneurons of the CA3 region of hippocampus as a basis for the study of large networks of CA3 neurons in which we can see how variations in key parameters can switch the network between normal and epileptiform activity. Sec. 8.3 used a number of models to explore the role of single cell dynamics versus emergent network properties in thalamocortical oscillations. In Sec. 9.2 we studied the simulation of a single Purkinje cell as a very detailed compartmental model with realistic ion conductances and synaptic currents in each compartment. Since this takes massive computing resources to simulate a single cell, the models of cerebellar function in Secs. 9.3 and 9.4 used simpler, single-compartmental models: but we pointed the way to future multi-level modeling which will relate system behavior to the fine details of neuronal function.

P6.5. Neurochemistry

Finally, we recall material assessing the biological grounding of learning rules used in the section on "Plasticity: Modeling Learning". Sec. 6.3 started by looking at one of the best-studied forms of dynamics at the synaptic level, namely long term potentiation (LTP), showing its implication in experimental studies of Hebbian synaptic modification, and analyzing models of potentiation based on AMPA and NMDA receptors. We linked this back to dynamics at the activity level by studying the role of NMDA receptors in the generation of oscillations at the cellular level. We also discussed the need for long-term depression (LTD) in Hebbian synapses.

Sec. 9.4 modeled the cerebellar role in saccade adaptation, extending our view of LTD by stressing the notion of a "window of eligibility" to constrain the timing relation between the parallel fiber "context" and the climbing fiber "training signal". Sec. 10.6 studied cortico-striatal plasticity, positing a form of reinforcement learning in which dopamine released by the SNc acts as the reinforcement signal to toggle between Hebbian learning (positive reinforcement, LTP) and anti-Hebbian learning (negative reinforcement, LTD). We also suggested that shaping of the eligibility signal may be task dependent, setting an important goal for neuroscience to bridge from this systems level of neural analysis to that of synaptic neurochemistry.

P7. The Hippocampus: a Case Study

It is generally agreed that the hippocampal formation has a crucial role in learning and memory processes. The hippocampus is reciprocally connected to many neural centers and it is thought to prepare information for long term storage. Moreover, the hippocampus has an important role in neurological diseases. Alzheimer’s disease, epilepsy and ischemia are associated with learning and memory impairment, and are accompanied by selective neuronal death or characteristic changes in the hippocampal circuitry. In this Section, we review the discussion of the hippocampus provided in Chapter 6 of Organization, and make some additional remarks in the light of recent results.

P7.1. Levels, Methods, Problems

The hippocampus has been studied on different levels and by different methods:

(i) the anatomical organization of the hippocampus including its afferent and efferent systems and the local circuitry of its components ;

(ii) the electrical activity patterns related to global brain states and the underlying single-cell activities;

(iii) the cellular synaptic plasticity that occurs during long term potentiation (LTP);

(iv) the role of the hippocampus of rats learning a spatial environment; and

(v) the function of hippocampus in human memory.

Computational theories try to understand how the hippocampal neural circuitry and the whole cortico-hippocampal loop, supplemented with specific subcortical inputs, can implement different types of dynamic activity ("brain states") such as theta rhythms and sharp waves, and how these activity patterns elicit long term potentiation (LTP). LTP is assumed to be the cellular basis for memory formation (Bliss and LØmo 1973). The relationship between the brain states and the enhancement of synaptic modifiability (LTP) can also be established by computational methods.

The functional view of the hippocampus related to navigation and memory phenomena should and could be unified with the structural approach by using dynamic computational models.

P7.2. Anatomical organization

P7.2.1 Global organization: The hippocampal formation is a cortical structure located in the temporal lobe. It is called archicortex for its evolutionary precedence over neocortex, and is relatively simple compared to neocortical structures. It has an elongated C-shaped form, and looks like a tube oriented perpendicular to the corpus callosum. Structurally the hippocampus is the simplest form of cortex, but this simplicity is in stark contrast with its role in processing information from the external world through the sensory systems, and from the "internal world" conveyed by subcortical inputs. Whereas, e.g., primary visual cortex is specialized for processing a single modality, the hippocampus is functionally one of the most complex supramodal association areas, with many routes to many cortical areas. Polymodal association areas converge directly or indirectly on the entorhinal cortex which in turn forms the principal source of afferents to hippocampus. The hippocampus receives refined information from virtually all sensory modalities, both exteroceptive and interoceptive, via entorhinal cortex, and is thought to prepare information for long term storage elsewhere in cortex with the return projections from hippocampus possibly providing cells in polymodal cortex with a "condensed sketch" of the overall context in which their unimodal input occurred.

P7.2.2. Cell types: The principal cells of the dentate gyrus, the granule cells , generally do not have basal dendrites, but only have spiny apical dendrites. Their axons form the mossy fibers, which pass through the hilus (the area contained within the formed by DG) before terminating on the dendrites of the CA3 pyramidal cells. The granule cell axons are considered to form excitatory synapses; the most likely neurotransmitter is glutamate. The hilus itself contains "polymorphic" cells, i.e., cells of varied morphology. The principal cells of the hippocampus proper, the pyramidal cells , have thick apical dendrites extending through the stratum radiatum up to the stratum lacunosum-moleculare, and shorter and thinner basal dendrites which arborize in the stratum oriens. The thick, myelinated main axons of the CA3 pyramidal cells arising from the soma and terminating in the stratum radiatum and oriens of the CA1 region are the Schaffer collaterals. Furthermore, CA3 pyramidal cells have recurrent collaterals terminating in the CA3 region itself. Axons of the CA1 pyramidal cells are thin, and provide part of the hippocampal output, projecting mostly to subiculum, and sometimes straight to the entorhinal cortex. Both CA3 and CA1 pyramidal cells also have collaterals which descend to the septal area via the fimbria. For most pyramidal neurons, glutamate is the (excitatory) neurotransmitter, which binds to (at least) three different receptor subtypes, metabotropic, AMPA and NMDA. Recently metabotropic excitatory amino acid receptors have also been taken into account. Recent data for the differential distribution of three types of glutamate receptor have been reviewed by Somogyi et al. (1998). Hippocampal "nonpyramidal" interneurons exhibiting local inhibitory effects have a decisive role in controlling electrical activity. Freund and Buzsáki (1996) reviewed the anatomical, neurochemical and pharmacological, cellular and system physiological data and showed the diversity of interneurons. Certain types of interneurons local control the activity of the principal cells, while others may form a network, and collectively exert the inhibitory effect. Interneuronal networks may exhibit network oscillations with different frequencies, and they control the synchronized operation of the principal cells and the formation of plasticity.

P7.2.3. Circuitry: According to the nowadays classical scenario (Andersen et al. 1971) there is a unidirectional cortico-hippocampo-cortical loop formed by the excitatory pathways. The perforant path originates in the entorhinal cortex and terminates in the granule cells of the dentate gyrus. The axons of the granule cells, i.e. the mossy fibers, project to the proximal part of the CA3 pyramidal cell dendrites. There is an extensive axonal arborization within the CA3 region. The axon collaterals of the CA3 pyramidal cells, i.e. the Schaffer collaterals, innervate the dendrites of the CA1 pyramidal cells, which further project to the subiculum and then to the entorhinal cortex. The anatomical organization is far more complex, having several other projections. The entorhinal cortex also innervates a subfield of the CA3 and CA1 regions. The local inhibitory cells can receive innervations from the principal cells of the same (feedback) or afferent (feed-forward) subfield. They can be innervated also by extrahippocampal afferents as well (Freund and Antal 1988, Freund et al. 1990, Gulyás et al. 1990). As we cited earlier (Somogyi et al. 1998), the basic principle of the organization of hippocampal circuitry seems to be known, even though many specific details are under clarification.

P7.2.4. Afferents and efferents: It is well established that the major input (the perforant path) to the hippocampus arises from layer II of entorhinal cortex (ENT). The ENT itself is considered as a relay for information coming from multimodal association areas in the temporal, prefrontal, cingulate, and insular regions. It seems likely that olfactory information is relayed through the lateral entorhinal cortex, while the medial entorhinal cortex conveys visual information. The former terminates in the outer third of the molecular layer, the postsynaptic targets being the distal dendritic field of the granule cells. The latter terminates on the middle third of the molecular layer. For a newer review of the morphological features of the entorhinal-hippocampal connections see Turner et al. 1998. Besides cortical (and commissural) connections different subcortical structures are identified as hippocampal afferents and efferents. Subcortical inputs, in general, may strongly modify the hippocampal activity patterns. After having a model of the cortico-hippocampo-cortical loop, the specific effects of different inputs can be studied.

Another type of fiber is found to be GABAergic and to exclusively innervate inhibitory interneurons of the hippocampus proper (Freund and Antal 1988, Gulyás et al. 1990). Since the interneurons contain mostly GABA, as transmitter substance, the GABA-GABAergic interaction implements the phenomenon called disinhibition. Though the number of fibers producing disinhibition is relatively low, still their modifying effects is strong. The quantitative details are not known, and extensive simulation experiments are necessary to discover them.

The raphé nuclei of the midbrain area innervate the hippocampus. The main neurotransmitter of the raphé-hippocampal projections is serotonin. Specifically, the median raphé-projections selectively innervate a subclass of interneurons in the CA regions (Freund et al. 1990), namely those containing calbindin, but not exclusively (Acsády et al. 1993).

One important output field of the hippocampus is the subiculum; other projections exist to the presubiculum, parasubiculum and the ENT. The subicular efferents to the deep layers of the entorhinal cortex close the multisynaptic entorhinal cortex-hippocampus-entorhinal cortex loop. Subiculum also generates a massive projection that travels in the fornix to the anterior thalamic nuclei and the mammillary bodies lying at the posterior edge of the hypothalamus. Deep layers of ENT are innervated by the hippocampus and project to neocortex, especially to zones neighboring ENT and to the medial frontal areas.

P7.3. Global Brain States and Behavioral States

P7.3.1. Electrical activity patterns: Global brain states, in both normal and pathological situations, may be associated with spontaneous activities of large populations of neurons. Experimentally, these activities may be detected by recording both from large neural assemblies (as in the EEG) or from a single neuron of the cell population. Generally, behavioral correlates can be defined for electrophysiologically global brain states.

Two main, normally occurring, global hippocampal states are known: the rhythmic slow activity, called the theta rhythm and the irregular sharp waves (SPW) (Buzsáki 1989). A pathological brain state, associated with epileptic seizures, the epileptiform patterns, is also characteristic of the hippocampus. More precisely, a set of different types of collective neural behaviors are qualified as "seizures". Both normal brain states and epileptic states are related to some (not clearly defined) synchronous activity. While a certain degree of synchronization is characteristic of normal rhythmic activity, highly synchronized cellular activity is more characteristic of clinical disorders. Other oscillations, such as a fast (40-100Hz) gamma oscillation found mostly in the hilus, and transient high-frequency (200Hz) oscillation in the CA1 region have also been reported.

P7.3.2. Theta rhythms: The theta rhythm is a population oscillation with large (~1 mV) amplitude and with 4-12Hz frequency. Originally, the theta rhythm was found to occur whenever the animal engages in such behaviors as walking, exploration, or sensory scanning, as well as in REM sleep. O'Keefe and Nadel (1978) suggested that displacement movements - but not stationary voluntary movements (e.g., bar pressing at low speeds) - in the rat, coincide with theta; moreover, the frequency of theta has been found to correlate with speed of movement (O'Keefe and Recce 1993). It can also be phase-locked to sensory stimuli. Buzsáki et al. (1994) speculated on the double functional role of hippocampal theta rhythm. First, a large-scale oscillation in the entorhinal-hippocampal network induced by the septum is maintained by phase-locking. Second, since the majority of the pyramidal cells are silent during theta, and their membrane voltage is kept close to but below the threshold, relatively few excitatory synapses are sufficient to discharge them. In addition, theta is involved in LTP generation.

P7.3.3. Sharp waves: Sharp waves (SPWs) have a very large amplitude (up to 3.5mV), their duration is 40-120ms, and their frequency can be between 0.2 and 5 Hz. Though maximal SPW frequencies do overlap theta frequencies, theta waves are much more regular than SPWs. SPWs also have behavioral correlates: they occur during awake immobility, drinking, eating, face washing, grooming and slow wave sleep. During SPWs, pyramidal and inhibitory cells fire with increased frequency. Furthermore, there is a partial synchronous cellular activity of both pyramidal and inhibitory neurons. The degree of synchrony is, however under the threshold for induction of epileptic seizure.

P7.3.4. Synchronization: While theta rhythms depend on septal input, SPWs are formed by internal processes. One important precondition for SPW generation is the occurrence of a population burst in a small set of CA3 pyramidal cells. Their synchronization is mediated by excitatory synaptic connections.

Epileptic activity occurs in a population of neurons when the membrane potentials of the neurons are "abnormally" synchronized. As we already know, a certain degree of synchrony is necessary for normal theta and SPW behavior, and the transition between normal and abnormal degrees of synchrony is not clear. Rather arbitrarily, activity has been considered epileptic if more than 25% of the cells fire during 100ms (Traub et al. 1992). In vitro models of epilepsy (Traub et al. 1987, Traub and Miles 1991, Traub and Miles 1992, Traub et al. 1992) offer a means to study the cellular mechanisms of the different types of epileptic phenomena by combined physiological and simulation methods.

Both experiments and theoretical studies suggest the existence of a general synchronization mechanism in the hippocampal CA3 region. Synaptic inhibition regulates the spread of firing of pyramidal neurons. Inhibition may be reduced by applying drugs to block (mostly) GABA-A receptors. If inhibition falls below a critical level, complete synchrony occurs. Collective properties of networks of pyramidal cells modulated by inhibition have been studied successfully by Traub and Miles (1991).

There are ongoing debates about the origin of cortical gamma oscillation. Gray and McCormick (1996) suggested that the source of the gamma frequency "chatter" may be the intrinsic property of the cell. Very recently Wang (1999) gave an ionic conductance model of chattering neurons (in the neocortex), where the backpropagation of the action potential from the soma to the dendrite, is a key element of the rhythm generation. Gamma oscillation, however, may be the network property of interneurons connected by GABA-A mediated inhibition. (Whittington et al. 1995), Wang and Buzsáki 1996, Traub et al. 1997.)

P7.3.5. Modeling rhythmic activity in the CA3 region of the hippocampus: Structure-based bottom-up modeling has two extreme alternatives, namely multi-compartmental simulations, and simulation of networks composed of simple elements. There is an obvious trade-off between these two modeling strategies. The first method is appropriate to describe the activity patterns of single cells, small and moderately large networks based on data on detailed morphology and kinetics of voltage- and calcium-dependent ion channels. The second offers a computationally efficient method for simulating large network of neurons where the details of single cell properties are neglected.

Traub and Miles (1991) simulate hippocampal (mostly CA3) population activity by building "bottom-up" models from data on anatomic connectivities, ionic conductances and synaptic properties. In most of their simulations the aim is to reproduce the results of physiological measurements made on hippocampal slices. Physiological measurements (both intracellular recording from one cell or, mostly, from a pair of cells, as well as field potential recording from a localized cell population) and simulations under various circumstances contribute to discovering the mechanism of both normal and pathological phenomena (e.g. epileptogenesis). Neurons in the Traub-Miles networks are modeled with a Hodgkin-Huxley formulation which has been modified in numerous way.

Two types of action potentials can be generated in the CA3 pyramidal cell: (i) fast, sodium-mediated, localized mostly to the soma, and (ii) slow, calcium-mediated, mostly in the apical dendrite. The role of the potassium channels is, roughly speaking, repolarization.

The response of CA3 pyramidal cells to injected currents, namely the intrinsic burst discharges, are reproduced by the model. The frequency, even the regularity, of the action potentials depends on the strength of the applied current. A burst consists of a series of fast spikes at intervals of 5-10ms terminating in one or more slower action potentials. The burst is called intrinsic, since isolated neurons can produce it. Some characteristic features of the physiological responses that were reproduced were (i) an intrinsic burst followed by a long after-hyperpolarization (AHP); (ii) the dependence of bursting on the resting potential; (iii) summation of spike afterdepolarization to produce a depolarizing envelope; and (iv) the ability to prevent full burst generation by properly timed hyperpolarizing input.

In Organization we offered only a few comments about population models. The description of large population of neurons requires a different methodological approach, namely the application of population theories. Just as collective phenomena emerging in physical systems made from large number of elementary components (spins, molecules, etc.) are treated by statistical mechanics, so, analogously, have statistical dynamic theories of neural populations been established (Wilson and Cowan 1973, Amari 1974, Ventriglia 1974, 1994). These neuronal population theories used oversimplified single-cell models. One important example is the lack of ability to generate burst mode.

In the last couple of years Érdi’s group has developed a population theory of bursting (and non-bursting) neurons (Érdi et al. 1997, Grobler et al. 1998, Barna et al. 1998) and applied it to simulating large-scale hippocampal activities. In this framework (i) the activity (different levels of subthreshold membrane potential/refractory state) distribution of groups of otherwise indistinguishable neurons is considered, and the subpopulations of neurons communicate via packets of impulses (action potentials) which they can emit and absorb; (ii) neurons and impulses (action potentials) form two distinct populations. (iii) The neurons, excitatory and inhibitory, occupy fixed positions in space, and their state is characterized by probability density functions over two continuous variables: their membrane potential and internal calcium concentration. (iv) Impulses can move from the point of emission (a neuron) to the point of absorption (another neuron) either by homogeneous spreading (random connectivity) or along prespecified paths (specific connectivity), carrying a quantum of excitation or inhibition (depending on the character of the emitting neuron). The absorption of impulses by a neuron implies: (A) change of the membrane potential; (B) firing of the neuron with a probability determined by the value of the membrane potential; (C) emission of new impulses as a result of firing.

Distribution functions for the probable number of (excitatory and inhibitory) impulses and neurons, and also for neurons in refractory state provide a statistical description of the system. To take into account the actual connectivity structure of the system, a set of absorption coefficients and emission coefficients are given. These values define the strength and efficacy of the excitatory and inhibitory effects at each point of the neural system. Further parameters incorporated into the model give the possibility of taking into account other specific biological details such as impulse generation from external source, spontaneous decay of subthreshold excitation, refractory period, synaptic delay etc. To evaluate and to visualize the simulation experiments, we use such macroscopic variables as the local density of impulses, the local mean net excitatory effect, and the local mean subthreshold excitation. Several normal epileptic activities, such as the synchronized population burst and synchronized synaptic potential (the analogue of SPW in slices) and the propagation of the stimulus, have been simulated, while the behavior of an "averaged" single neuron was also shown.

P7.4. Brain States and Long Term Potentiation

Long-term potentiation (LTP) was first discovered in the hippocampus and is very prominent there. LTP is an increase in synaptic strength that can be rapidly induced by brief periods of synaptic stimulation and which has been reported to last for hours in vitro , and for days and weeks in vivo . This time-course may be insufficient to sustain long-term memory, but there appear to be multiple LTP mechanisms, and one dependent on protein synthesis might serve long-term memory: inhibition of protein synthesis disrupts the maintenance of LTP, but leaves the induction of LTP relatively or totally intact. It is possible to relate properties and mechanisms of long-term synaptic plasticity in the mammalian brain to learning and memory.

There is now evidence for both homosynaptic and heterosynaptic LTP in area CA1 of the hippocampus, and an associative form of LTP has been reported in hippocampal CA1 and dentate gyrus. Hebbian synaptic modification depends on the co-occurrence of pre- and postsynaptic activity, and this effect was found in the form of LTP occurring in the Schaffer collateral/commissural synaptic input to the pyramidal neurons of hippocampal CA1 (Bliss and Collingridge, 1993).

Buzsáki et al. (1994, pp 168) argue that sequential potentiation mechanisms "ensure that discharge of a given set of entorhinal neurons during subsequent visits to the same part of [a] maze (recall) will reactivate the same subsets of neurons in CA3 and CA1. The hierarchy of neuronal firing during the SPW-associated bursts, therefore, is precisely determined by the recent past of the neural network. The rules of burst initiation and reconvergent excitation, subserved by the anatomical-physiological organization of the CA3 region, ensure that the synchronized events during consummatory behaviors and slow wave sleep carry biologically meaningful information..."

P7.5. Hippocampal Function, Cognitive Maps, Human Memory

The functional view of the hippocampus - its role in the cognitive maps underlying navigation and spatial behavior in rats - should be combined with a dynamic view of synaptic plasticity, seeing how Hebbian-like plasticity may enable hippocampal cells to learn to encode different "places" in a cognitive map.

P7.5.1. Place cells, navigation: Rats are highly exploratory. In a new environment, they tend first to explore outward from some base, then to shift to other bases until they become highly adept at navigating from one place to another, visiting sites where food has been taken, and returning to inaccessible hiding places. Rats entering one arm of a T-maze will tend to choose the other arm on the next exposure ("spontaneous alternation"). A landmark is not merely a stimulus to be approached for a reward. Rats remember headings relative to the landmark, and can use the position of a number of objects to navigate towards, e.g., a food source or hiding place. In the "water maze" (Morris 1981), a rat can use such cues to swim to a platform located beneath opaque water.

Certain pyramidal cells of the CA1 and CA3 region fires when the rat moves to a particular place in the environment, and these cell are called, as "place cells". There are some suggestions for how learning leads to the appearance of stable, bounded place fields as a result of exploratory behavior. Jensen and Lisman (1996) and Wallenstein and Hasselmo (1997) assume that the dominating factor of the information exploited during learning is consistency in the sequence of perceived sensory input. It is assumed that the important thing is not what the rat actually sees, smells or touches, but that a particular input pattern is always followed by a (different) particular input pattern. According to this idea, place cells are sensitive to sub-sequences in the whole input stream, which we observe as spatial sensitivity due to the spatiotemporal continuity of the rat's movement. Models based on this temporal correlation in the input stream can reproduce many important features of place fields. However, they cannot account for the symmetric graded firing profile within a place field. Recently, we gave a learning-rule based model (Érdi et al. 1998; Szatmáry et al., submitted) to overcome this failure of previous models.

Experiments suggest that rats (i) have associative memory for complex stimulus configurations, (ii) can encode the spatial effect of their own movements, and (iii) are able to form sequences of actions to go from a starting location to a goal. In other words they have cognitive map (Tolman 1932). The hippocampus may function as part of a local navigation system (i.e., cognitive map).

Several neurobiologically (more or less) realistic models and algorithms have been suggested to solve the problem of orientation and navigation based on information obtained from place cell firing (Burgess et al. 1994, Burgess and O'Keefe 1996, Gerstner and Abbott 1997, Zhang et al. 1998). Burgess et al. (1994) constructed a multi-layered network of different functionally defined cells (entorhinal, place, subicular, head-direction, and goal cells) supplemented with layer-specific activity and learning dynamics. Relevant neurophysiological phenomena (theta rhythm, phase coding, place fields) are incorporated into the model.

P7.5.2. Affordances, Motivation, and the World Graph Theory: O’Keefe and Nadel (1978) distinguish two paradigms for navigation, the "locale system" for map-based navigation and the "taxon (behavioral orientation) system" for route navigation. Guazzelli, Corbacho, Bota, and Arbib (1998) model both the taxon system and the map-based system, as well as their interaction; and they argue that the map-based system involves the interaction of hippocampus with other systems, not just the hippocampus alone. They relate taxis (movement towards some goal, as in phototaxis) to the notion of an affordance (a visual indication of a course of action, Gibson 1966; already presented for grasping in Sec. P5.4). Just as a rat may have basic taxes (plural of taxis) for approaching food or avoiding a bright light, so does it have a wider repertoire of affordances for possible actions associated with immediate sensing of its environment. We propose that affordances are extracted by the rat posterior parietal cortex, which guides action selection by the premotor cortex and is also influenced by hypothalamic drive information. The Taxon-Affordances Model (TAM) for taxon-based determination of movement direction is based on models of frog detour behavior, with expectations of future reward implemented using reinforcement learning. The specification of the direction of movement is refined by current affordances and motivational information to yield an appropriate course of action. The World Graph (WG) theory expands the idea of a map by developing the hypothesis that cognitive and motivational states interact. Guazzelli et al. (1998) developed an integrated TAM-WG model which explains data on the behavior of rats with and without fornix lesions which disconnect the hippocampus from other neural systems.

P7.5.3. Memory systems: The role of hippocampal function in human memory is discussed in Organization by introducing the crucial dichotomies of procedural vs. declarative memory and of skill vs. episodic learning. The data suggest that the hippocampus is involved in declarative rather than procedural memory and in episodic rather than skill learning. It is also likely that hippocampus may form but not store memory traces.

There are ongoing debates on the role of hippocampus in episodic and declarative memories. Varga-Khadem et al. (1997, Mishkin et al. 1998) came out with the novel idea that hippocampus might play a selective role in episodic memory, while the related cortical structures might support semantic memory even in the absence of hippocampal function. While Tulving and Markowitsch (1998) support the new proposal, Squire and Zola (1998) do not see sufficient proofs that episodic and semantic memory are differently affected in amnesia.

Although the hippocampus stores information, it seems that it also "installs" this processed information elsewhere in cerebral cortex for long-term availability. Buzsáki (1989) suggested an informal model of memory formation in which cortical information is processed by two stages. First, during the theta brain state, cortical activity weakly potentiates, via the granule cells, the CA3 pyramidal cells associated to a labile form of memory trace. This weak potentiation initiates population bursts implying a transition from theta to SPW state. Under the SPW state, excitatory synapses between pyramidal cells both within the CA3 region and between CA3 and CA1 regions are enhanced. These enhanced synapses would be the substrate of a long-lasting memory trace. Since SPW and associated high-frequency oscillation in the CA1 region yields discharge of neurons of deep layers of entorhinal cortex and, it seems to be likely that hippocampal output may affect other neocortical targets, thus transferring information stored temporarily in the CA3 region to the neocortex for long-term storage.

This transfer may occur, at least in part, during sleep. Pavlides and Winson (1989) showed that hippocampal cells active during a waking period exhibit increased firing rates in the following sleep period. To investigate this effect in more detail, Wilson and McNaughton (1994) monitored the simultaneous activity of 50 to 100 CA1 cells during a running period (RUN) and during both the pre-behavioral (PRE) and post-behavioral (POST) sleep periods. During the RUN period, cells with overlapping place fields exhibited highly correlated activity; those with non-overlapping fields did not. Indeed, cells that were coactive during the RUN period showed a far greater correlation than during the PRE period. Moreover, this correlation was reactivated during the POST period, but declined with a time constant of approximately 12 minutes. Wilson and McNaughton see this as support for the hypothesis that hippocampal activity during sleep exhibits a reactivation of population activity from the prior waking period. Since CA1 has little direct connectivity between pyramidal cells, they suggest the correlations arise in CA3 (which has many intrinsic connections) or entorhinal cortex.

In support of the idea that information is transferred from hippocampus to neocortex especially during the synchronized bursts ("ripples") of sharp wave (SPW) activity (Buzsáki 1989), Wilson and McNaughton (1994) found during the POST period that correlations during ripples were significantly greater than the correlations in the periods between ripples. Chrobak and Buzsáki (1996) have shown that SPWs and ripples are initiated in CA3 and that the output layers - but not the input layers - of entorhinal cortex exhibit neuronal activity correlated with CA1 SPWs. Wilson and McNaughton thus suggest that the induced correlations during SPWs arise from modifications within the hippocampus and are propagated to the output layers of entorhinal cortex.

P7.5.4. Functional imaging of the Human Hippocampus: The last few years have seen a tremendous development in brain imaging, including that related to hippocampus (see, e.g., the recent issue of Hippocampus 9(1), 1999). Stern and Hasselmo (1999) integrated cellular and fMRI studies, while Horwitz et al. (1999) showed that there only very few large-scale neuronal models to relate PET data to neuronal activity. Much remains to be done. Functional Imaging is certainly underrepresented in Organization. However, we offer our own work on synthetic PET (Arbib et al. 1995) and the population models developed in Érdi’s group (Grobler et al. 1998, Barna et al. 1998) as two steps toward a technique to understand the results of brain imaging in terms of detailed neural activity.

P8. Towards a Cognitive Neuroscience

Our studies of structure, function, and dynamics are located in a broad sweep which runs all the way from the motility of the embryo to the learning of visually guided behavior. In this concluding section we turn from retrospect to prospect, suggesting ways in which the ideas developed in Organization may contribute to future work in cognitive neuroscience. This prospectus embodies a strong philosophical position, namely that mind (at least that aspect of it known as cognition) can be explained in terms of the workings of matter (especially that structured as neural systems). This raises a methodological challenge since the categories of "mind-talk" – function - and "brain talk" – structure - do not map directly one on to the other. Organization has provided a framework in which the study of structure and function may be integrated with dynamics. In doing so, we have reflected on the immense progress neuroscience has made in delimiting structure, whether the functional neuroanatomy that has, for example, used double labeling techniques to subdivide and chart the terra hitherto incognita of the primate association cortices, or the studies in neurochemistry and molecular neurobiology that reveal more and finer structures within the individual neuron. We have also seen conceptual advances in the study of large networks of (somewhat simplified) neurons, ranging from studies of low-level vision to the statistical mechanics of self-organization which emphasize the matching of a single function to a single network. Chapter 3 of Organization emphasized work at a different level, in which a network of functions (schemas and schema assemblages) must be mapped to a network of neural networks. We now discuss the implications of this for cognitive neuroscience.

P8.1. Memory, Perception, and Intelligence

Among the properties that contribute importantly to intelligence are the following:

Possession of a modifiable model of the world, with its attendant adaptability : A system to act intelligently must not only be able to take properties of its environment into account, but must be able to update its record of these properties to take account of new observations and changing relationships.

Flexibility and generality: An intelligent system must not only use past experience to act adaptively, but must also be able to apply its past experience to situations which are not superficially similar to those encountered before. Again, techniques which have been developed to solve one type of problem should be recognized as applicable even when a very different domain of problems is involved.

Dynamic Planning: An intelligent system should use its model to plan and evaluate alternative courses of action before committing itself to one of them. For a symbol-manipulation system there may be little real distinction between planning and action, but for a robot or an animal the distinction is very real and very important it pays to recognize a precipice in advance and plan to avoid it rather than recognizing one's mistake after going over. However, it is crucial that the plan be dynamic in that it be rapidly and effectively updatable when new data reveal unexpected obstacles or make sought-for information available.

One important form of working memory is obtained by holding a particular pattern of firing during a delay task. Such neurons have been found in dorsolateral prefrontal cortex as well as hippocampus. What distinguishes these two systems? The answer is still far from clear, but we assert that a full analysis of the procedural/declarative distinction in humans requires a theory of consciousness which distinguishes conscious/declarative from non-conscious/procedural access to schemas. In schema-theoretic terms: the perception of a situation or the carrying out of a particular action requires the construction of a particular schema-assemblage (assimilation, in Piaget's terms); to the extent that the perception or action is problematic, the schemas may become modified to increase the chance of success in similar situations in future (accommodation). We learn both by storing schema-assemblages (memory of a specific situation and course of behavior) and by tuning extant schemas. However, it is too facile to say that the former corresponds to "fact memory", and the latter to "skill memory", since "skill memory", too, partakes of some aspects of assemblage-formation, since skills are tuned versions of "programs" constructed from prior schemas, rather than only by tuning of single previously extant. The fact that a schema may be activated without conscious awareness emphasizes the notion that different neural processes must be involved in monitoring the use of a schema as distinct from the use per se of the schema. The "what"/"how" distinction reviewed in Chapter 8 - one patient may be able to "declare" the size of an object yet not be able to preshape the hand appropriately to grasp it; another patient may exhibit the opposite - shows that some schemas are instantiated on paths to conscious awareness, and others are not. Moreover, some at least of the "working memory" systems of prefrontal cortex are tightly coupled to specialized areas of parietal cortex, and are thus tightly integrated into the procedural "how" system rather than the conscious/declarative "what" system. So: the loop of explanation must be closed back from the theory of consciousness. This is consistent with Rozin's (1976) view that procedural learning may be phylogenetically old, having developed as a collection of encapsulated special-purpose abilities of specific neural systems to register cumulative changes in their functioning. By contrast, the capacity for declarative learning reaches its full development only with the elaboration of medial temporal areas in mammals, especially hippocampus and related cortical areas. Organization offers some relevant material in the discussion of Hippocampal Function and Human Memory in Sec. 6.5.

Perception provides access to motor schemas to control interaction with the object, but this does not necessarily entail execution of even one of these motor schemas. Although an animal may perceive many aspects of its environment, only a few of these can at any time become primary loci of interaction. Planning is the process whereby the system combines an array of relevant knowledge to determine a course of action suited to current goals. In its fullest subtlety, planning can involve the refinement of knowledge structures and goal structures, as well as action per se . Novel inputs (e.g. coming upon an unexpected obstacle) can alter the elaboration of high-level structures into lower-level tests and actions which in turn call upon the interaction of motor and sensory systems (cf. the notion of dynamic planning). We seek to study programs which are part of the internal state of the system, and which can flexibly guide ongoing action in terms of internal goals or drives and external circumstances. Note that we do not imply that planning is a conscious process, and that planning goes beyond mere choice. In "choice", we suggest that a decision (whether conscious or not) must be made between a few clearly delimited alternatives. In planning, by contrast, solutions to many, possibly conflicting, subproblems will have to be constructed to yield a, possibly quite novel, course of action. Organization provides an evolutionary view of how visual perception may be evolve into a distributed capability for planning in Sec. 8.6, From Action-Oriented Perception to Cognition.

An important gap in most computational analyses of the mind comes about because few neuroscientists think about the social nature of being a human (see Brothers and Ring 1992, 1993 for an entry point to that small literature which does begin to link neural activity to social cognition). To be human is not just to have a sophisticated "computer" in the head called the brain. It is also to have grown up as a member of society, and to have learned the nuances of that society. Neuroscientists and cognitive scientists emphasize what they can measure objectively such as language where we analyze a string of symbols, or vision where there are particular patterns to which we can see how people or animals or neurons respond. Arbib's (1985) In Search of the Person emphasized that much of human experience or, if you will, Person-reality, of being a member of society, being aware, and having experience of love, hate and anguish is normally not addressed at all within the framework of brain research or cognitive science. The point was not to reduce these to current brain theory or cognitive science, but rather to show how the science and the personal experience might be thought about in a unified framework in which understanding of each reality could come to shape that of the other.

P8.2. Neural-Cognitive Interaction: Self-Organization, Constructivism, Downward Causation, Hermeneutics

Our intention was to write a "neural book" and to show brain components are organized to implement "higher brain functioning". Still, we cannot avoid a brief mention of some philosophical issues. John Szentágothai (1993) argued that the essence of the neural is to be found in its self-organizing character. If all neural functions, including those from the simplest elementary reflexes to complex global functions of the whole organism have at their very basis spontaneous activities arising - in part at random, in part constrained by the genome - in individual nerve cells and if all neural functions are integrated by self-organization into various activity patterns, our whole understanding of the neural organization has to undergo rather fundamental changes. If the reflex paradigm of neural systems is to be abandoned for the new concept of "self-organization" of spontaneous (random or other) activity, this would be an entirely new challenge for "brain-mind philosophy".

In a recent BBS target article, the idea of constructionism (Quartz and Sejnowski, 1997) emerged, providing arguments for the necessity of the interaction between cognitive and neural levels of description. The statement, that cognitive level learning influences brain development may be associated to the argument to the problem of "downward causation": Downward causation, i.e. the notion that mental agents can influence neural functioning, was suggested by Sperry (1969). It was not clear, however, how conscious processes directly influence physiological mechanisms in Sperry’s scheme. Szentágothai (1984) was more cautious. He suggested that the nervous system can be considered as being open to various kinds of information, and that there would be no valid scientific reason to deny the existence of downward causation, more precisely, a two-way causal relationship between brain and mind. Moreover, the grand tradition of philosophical hermeneutics says that we can learn and understand what we already know and understand in embryological form.

Érdi (1996) argues that the philosophical tradition of hermeneutics, i.e., the "art of interpretation", which is a priori neither monist nor dualist, can be applied to the brain. Playing with the idea that the "device approach" to the brain and the philosophical approach can be reconciled, he concluded that the brain is a physical structure which is controlled and also controls, learns and teaches, process and creates information, recognizes and generates patterns, organizes its environment and is organized by it. It is an "object" of interpretation, but also it is itself an interpreter. The brain not only perceives but also creates new reality: it as a hermeneutic device. . In The Metaphorical Brain 2 , Arbib (1989) argues that our theories of the brain are metaphors, while the brain itself represents the world through schemas, which may themselves be viewed as metaphors (and see Arbib and Hesse 1986 for the role of hermeneutics in a schema-based epistemology that links the individual and the social, especially the notion of two-way reduction ).

P8.3. Language

To end on a more concrete note, we discuss briefly the controversial issue of whether or not the study of language can be illuminated by approaches to neural organization of the kind presented in Organization.

Connectionist modelers distinguish two main architectures for their networks. In a layered feedforward network, the "neurons" are arranged in a series of layers, with the only connections being from neurons in one layer to neurons in the next layer. Since there are no loops, there is no possibility of a "reverberating memory", and thus, after a suitable propagation delay, each input pattern yields a unique output pattern. By contrast, a network with loops acts as a dynamic system: for each fixed input pattern, patterns of activity may move around the network, creating dynamic sequences of internal states. In many studies, the behavior of such a network is not characterized in terms of input-output pairs, but rather in terms of settling into a "steady state" such as a point attractor , a limit cycle (yielding a sustained oscillation) or even a strange attractor (deterministic chaos). Networks studied from this point of view are thus often referred to as attractor networks (cf. Organization, Sec. 4.5.4, Computation with Attractors: Scope, Limits and Extensions). Studies of lesioned attractor networks (e.g., Hinton and Shallice 1991; Plaut and Shallice 1993) provide further insights into the clinical observations of neuropsychology. Simulated damage to attractor networks can qualitatively mimic some of the deficits seen following human brain damage. In particular, such studies suggest how there may be an appearance of functional modularity (i.e., two functions may be differentially impaired by network damage) even when the functions are implemented by a single network.

Despite its real contributions, such studies make little progress in explaining the contributions of specific brain regions to language capabilities. Recent work on "mirror" neurons (di Pellegrino, Fadiga, Fogassi, Gallese and Rizzolatti 1992) provides promise of in this direction by suggesting a new path for integration between the study of human language and the study of detailed neural mechanisms of visuomotor coordination. These mirror neurons are a subset of the grasp-related neurons of the F5 region of premotor cortex (Sec. 8.4.3). They discharge when the monkey observes meaningful hand movements made by the experimenter, such as placing or taking away objects from a table, grasping food from another experimenter, or manipulating objects. There is always a link between the effective observed movement for a mirror neuron and the effective executed movement.

These data suggest that area F5 is endowed with an observation/execution matching system and led Rizzolatti, Fadiga, Matelli, Bettinardi, Perani, and Fazio (1996) to seek evidence for an observation/execution matching system in humans. In a PET study of brain activation of humans observing hand gestures, they found a highly significant activation of the posterior part of the left inferior frontal gyrus - the rostral part of Broca's area! While homologies between cortical areas of different species are always difficult, a good case can be made that Broca's area is in part homologous with F5. These data led Rizzolatti, Fadiga, Gallese, and Fogassi (1995) to a bold hypothesis: namely that the functional specialization of human Broca's area derives from an ancient mechanism related to production and understanding of motor acts. To this we would add that this specialization may correspond to verbs or verb phrases, but seems separate from the functions of naming and of noun phrases more generally - concordant with our emphasis that language, like other functions studied in this volume, is to be seen more as a distributed function ("cooperative computation") rather than being a "unitary faculty". Rizzolatti et al. argue more specifically that the sophisticated capacity of action analysis shown by mirror cells is at the basis of the evolutionary prevalence of the lateral motor system over the medial (emotion-related) one in becoming the main communication channel in higher primates and man. Much work is currently underway (e.g., Rizzolatti and Arbib, 1998) to turn this hypothesis into a rigorous neurolinguistic model subject to coherent testing that integrates monkey neurophysiology and human brain mapping within a framework offered by the current debate over language evolution (Pinker and Bloom 1990, Wilkins and Wakefield 1995).

The point for our current claim - that Organization provides powerful tools for Cognitive Neuroscience - is that we see here an approach to language which does not treat it in grand isolation in the style of Chomsky, but instead (without denying the special character of these Higher Mental Functions) sees language and other cognitive processes within the framework of Neural Organization in general that we have charted with John Szentágothai.

Acknowledgment: This work was supported by the OTKA grants T025472 and T25500 (P.É.), and by the Human Brain Project under P20 Program Project Grant HBP: 5-P20-52194 (M.A.A.).

References

Acsády, L. Halasy, K., Freund TF, 1993, Calretinin is present in non-pyramidal cells of the rat hippocampus - III. Their inputs from the median raphe and medial septal, Neuroscience 52:829-841.

Adrian, E.D., 1942, Olfactory reactions in the brain of hedgehog, J. Physiol. 10, 459-473.

Adrian, E.D., 1950, Sensory discrimination with some recent evidence from the olfactory organ, Br. Med. Bull. 6:330-331.

Amari, S., 1974, A method of statistical neurodynamics, Kybernetik 14:201-215

Andersen, P., Bliss, T.V.P. and Skrede, K.K., 1971, Lamellar organization of hippocampal excitatory pathways. Exp. Brain Res. 113: 222-238

Anderson, J.R., 1983, The Architecture of Cognition Harvard University Press

Aradi, I., Barna, G., & Érdi, P. and Grõbler, T., 1995, Chaos and learning in the olfactory bulb, Int. J. Intelligent Systems, 10: 89-117

Aradi, I. and Érdi, P., 1996, Multicompartmental modeling of neural circuits in the olfactory bulb. Int. J. Neural Syst. 7:519-527.

Arbib, M.A., 1985, In Search of the Person: Philosophical Explorations in Cognitive Science, Amherst: University of Massachusetts Press.

Arbib MA (1989) The Metaphorical Brain 2: Neural Networks and Beyond, New York: Wiley-Interscience.

Arbib, M.A., Bischoff, A., Fagg, A.H., and Grafton, S. T., 1994, Synthetic PET: Analyzing Large-Scale Properties of Neural Networks, Human Brain Mapping, 2:225-233

Arbib MA, Caplan D (1979): Neurolinguistics must be computational. Behav. Brain Sci 2:449-483.

Arbib, M.A., Caplan, D., and Marshall, J.C. (1982): Neurolinguistics in Historical Perspective, in Neural Models of Language Processes (M.A. Arbib, D. Caplan, and J.C. Marshall, Eds.), New York: Academic Press, pp. 5-24.

Arbib, M.A., Conklin, E.J., and Hill, J.C., 1987, From Schema Theory to Language, Oxford University Press.

Arbib, M.A., and Hesse, M.B., 1986, The Construction of Reality, Cambridge University Press.

Arbib, M.A. and Hill, J.C., 1988, Language Acquisition: Schemas Replace Universal Grammar, in Explaining Language Universals, J.A. Hawkins, Ed., Basil Blackwell, pp. 56-72.

Arbib, M.A., Érdi, P. and Szentágothai, J., 1997, Neural Organization: Structure, Function, and Dynamics, Cambridge, MA: The MIT Press.

Arbib, M.A., Bischoff, A., Fagg, A.H., and Grafton, S. T., 1994, Synthetic PET: Analyzing Large-Scale Properties of Neural Networks, Human Brain Mapping, 2:225-233.

Barna G, Grôbler T, and Érdi P: Statistical model of the hippocampal CA3 region II. The population framework: model of rhythmic activity in the CA3 slice. Biol. Cybernetics, 79 (309-321) 1998

Bhalla, H.S. and Bower, J, 1993, Exploring parameter space in detailed single neuron models: simulations of the mitral and granule cells of the olfactory bulb, J. Neurophysiol. 69: 1948-1965.

Biedenbach M.A., 1966, Effects of anesthetics and cholinergic drugs on prepyriform electrical activity in cats, Exp. Neurol., 16: 464-479

Bliss, T. V. P., and Collingridge, G. L., 1993, A synaptic model of memory: Long-term potentiation in the hippocampus, Nature, 361: 31-39.

Bliss, T.V.P. and Lømo, T., 1973, Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of perforant path, J. Physiol. 232: 331-356.

Bradley DC, Garrett SL, Zurif EB (1980): Syntactic deficits in Broca's aphasia, in Biological Studies of Mental Capacities, Caplan D, Ed., Cambridge, MA: The MIT Press.

Britten KH, 1998, Clustering of response selectivity in the medial superior temporal area of extrastriate cortex in the macaque monkey. Vis Neurosci 1998 15:553-558.

Broca, P. (1861): Nouvelle observation d'aphémie produite par une lésion de la moitié posterieure des deuxième et troisième circonvolutions frontales, Bulletin de la Société Anatomique, 6:398-407.

Brothers, L. and Ring, B.,1992, A Neuroethological Framework for the Representation of Minds, J. Cog. Neuroscience, 4:107-118.

Brothers, L. and Ring, B., 1993, Mesial temporal neurons in the macaque monkey with responses selective for aspects of social stimuli, Behav. Brain Res. 57:53-61.

Burgess, N., Recce, M. and O'Keefe, J., 1994, A model of hippocampal function, Neural Networks, 7: 1065-1081

Burgess, N. and O'Keefe, J., 1996, Neuronal computations underlying the firing of place cells and their role in navigation. Hippocampus, 6:749-62

Buzsáki, G., 1989, Two-stage model of memory trace formation: a role for "noisy" brain states, Neuroscience 31: 551-570

Buzsáki G, Bragin A, Chrobak JJ, Nádasdy, Z., Sík, A., Hsu, M. and Ylinen, A., 1994, Oscillatory and intermittent synchrony in the hippocampus: relevance to memory trace formation, in Temporal Coding in the Brain, (Buzsáki G, Llinás R, Singer W, Berthoz A, Chrisen Y, eds.), Springer-Verlag, Berlin, pp. 145-172

Caplan, D. (1995): The Cognitive Neuroscience of Syntactic Processing, in The Cognitive Neurosciences, (M. Gazzaniga, Ed.), Cambridge, MA: A Bradford Book/The MIT Press, pp.871-879.

Castiello, U., Paulignan, Y. & Jeannerod, M. 1991 Temporal dissociation of motor responses and subjective awareness: A study in normal subjects. Brain. 114, 2639-2655

Chomsky, N. (1965): Aspects of the Theory of Syntax, Cambridge, MA: The MIT Press.

Chrobak, J.J., and Buzsáki, G. (1996) High-frequency oscillations in the output networks of the hippocampus-entorhinal axis of the freely-behaving rat', J. Neurosci. 16:3056-3066

Collins A.M and Quillian M.R., 1969, A spreading-activation theory of semantic processing. Psychol. Rev. 82,45-73.

Collins, A.M., and Loftus, E.F., 1975, A spreading activation theory of cognitive processing, Psychological Review, 82:407-42

di Pellegrino, G., Fadiga, L., Fogassi, L., Gallese, V., and Rizzolatti, G., 1992, Understanding motor events: a neurophysiological study. Exp. Brain Res. 91:176-180.

Duchamp-Viret P., Duchamp A. and Chaput M., 1993, GABAergic control of odor-induced activity in the frog olfactory bulb: electrophysiological study with picrotoxin and bicuculline, Neuroscience 53:111-120.

Érdi, P, 1996. The brain as a hermeneutic device. BioSystems 38: 179-189.

Érdi P., Grõbler T., Barna G. and Kaski K., 1993, Dynamics of the olfactory bulb: bifurcations, learning, and memory, Biol. Cybernetics, 69: 57-66.

Érdi P, Aradi I, Grôbler T, 1997, Rhythmogenesis in single cells and population models: olfactory bulb and hippocampus. BioSystems, 40:45-53.

Érdi P, Kepecs Á, Lengyel M, Obermayer K, Szatmáry Z, 1998, Dynamics of the Hippocampus: Multiple Strategies. International Conference on Neural Information Processing, Kitakyushu, (777-780)

Fagg, A. H., and Arbib, M. A., 1998, Modeling Parietal-Premotor Interactions in Primate Control of Grasping, Neural Networks, 11:1277-1303.

Frazier, L. (1990): Exploring the architecture of the language-processing system, in Cognitive Models of Speech Processing: Psycholinguistic and Computational Perspectives, G.T.M. Altmann, Ed., Cambridge, MA: A Bradford Book/The MIT Press, pp.409-433.

Freeman, W.J., 1975, Mass action in the nervous system, New York: Academic Press.

Freeman W.J., 1978, Spatial properties of an EEG event in the olfactory bulb and cortex, Elect. Clin. Neurophys. 44: 585-605.

Freeman, W.J., 1991, Nonlinear dynamics in olfactory information processing, in: Olfaction. A model system for computational neuroscience (Davis J.L. and Eichenbaum H., eds.), The MIT Press, Cambridge, MA, pp. 225-249

Freeman, W.J. and Schneider W., 1982, Changes in spatial patterns of rabbit olfactory EEG with conditioning to odors, Psychophysiology 19: 44-56.

Freund, T.F., Antal, M., 1988a, Septal GABAergic control of the hippocampus: a novel mechanism, Nature, 336: 170-173.

Freund and Buzsáki (1996) Interneurons of the Hippocampus, Hippocampus 6:347-470.

Freund, T.F., Gulyás, A.I., Acsádi, L., Görcs, T. and Tóth, K., 1990, Serotonergic control of the hippocampus via local inhibitory interneurons, Proc. Natl. Acad. Sci., USA 87:8501-8505.

Fulton, J.F. (Ed.), 1938, Physiology of the nervous system, London: Oxford University Press.

Gerstner W. and Abbott, LF, 1997, Learning navigational maps through potentiation and modulation of hippocampal place cells. Journal of Computational Neuroscience, 4: 79-94

Gibson, J. J. (1966). The Senses Considered as Perceptual Systems. Allen and Unwin.

Goodale, M. A., and Milner, A. D., 1992, Separate visual pathways for perception and action, Trends in Neuroscience, 15:20-25.

Grafton, S. T., Fagg, A. H., & Arbib, M. A., 1998, Dorsal Premotor Cortex and Conditional Movement Selection: A PET Functional Mapping Study. Journal of Neurophysiology 79:1092-1097

Gray ChM and McCormick DA, 1996, Chattering cells: superficial pyramidal neurons contributing to the generation of synchronous oscillations in the visual cortex. Science 274:109-113.

Grôbler T, Barna G, and Érdi P: Statistical model of the hippocampal CA3 region I. The single-cell module: bursting model of the pyramidal Cell. Biol. Cybernetics, 79 (301-308) 1998

Grôbler, T., Marton, P., and Érdi, P., 1991, On the dynamic organization of memory. A mathematical model of associative free recall, Biol. Cybernetics 65:73-79

Guazzelli, A., Corbacho, F.J., Bota, M., and Arbib, M.A., 1998, Affordances, Motivation, and the World Graph Theory, Adaptive Behavior, 6:435-471

Gulyás, A.I., Görcs, T.J. and Freund, T.F., 1990, Innervation of different peptide-containing neurons in the hippocampus by GABAergic septal afferents. Neuroscience 37:31-44.

Hinton, G. E. and Shallice, T. (1991): Lesioning an attractor network: Investigations of acquired dyslexia. Psychological Review, 98:74-95.

Jensen, O. and Lisman, J.E., 1996, Hippocampal CA3 region predicts memory sequences: accounting for the past precession of place cells. Learning and Memory, 3: 279-287

Hubel, D.H. and Wiesel, T.N., 1959, Receptive fields of single neurones in the cat's striate cortex, Journ. of Physiol, 1481:574-591

Lev DL and White EL, 1997, Organization of pyramidal cell apical dendrites and composition of dendritic clusters in the mouse: emphasis on primary motor cortex, Eur J Neurosci. 1997 9:280-90.

Li, Z. and Hopfield, J.J., 1989, Modeling the olfactory bulb and its neural oscillatory processing, Biol. Cybernetics 61:379-392.

Lichtheim, L. (1885) On Aphasia, Brain, 7:433-484

Lorente de Nó, 1938, Analysis of the activity of chain of internuncial neurones, J. Neurophysiol. 1: 207-244.

Luria, A.R., 1973,The Working Brain. Penguin Books.

Matelli, M., Luppino, G. & Rizzolatti, G. 1985 Patterns of cytochrome oxidase activity in the frontal agranular cortex of macaque monkey. Behav Brain Res. 18, 125-137

McDonald, J., and MacWhinney, B. (1989): Maximum likelihood models for sentence processing, in A Cross-Linguistic Study of Sentence Processing, (B. MacWhinney and E. Bates, Eds.), Cambridge: Cambridge University Press, pp.397-422.

Mishkin, M., Ungerleider, L. G., and Macko, K. A., 1983, Object vision and spatial vision: Two cortical pathways, Trends in Neuroscience, 6: 414-417

Mishkin, M. Varga-Khadem, F and Gaian, N 1998, Amnesia and the Organization of the Hippocampal System. Hippocampus 8:212-216.

Morris, R.G.M., 1984, Developments of a water-maze procedure for studying spatial learning in the rat. Journal of Neuroscience Methods, 11:47-60

Mountcastle, V.B., 1957, Modalities and typographic properties of single neurones of the cat's sensory cortex, J.Neurophysiol., 20: 408-434

Mountcastle VB, The columnar organization of the neocortex, 1997. Brain 120:701-22

Muakkassa, K. F. & Strick, P. L. 1979 Frontal lobe inputs to primate motor cortex: evidence for four somatotopically organized "premotor" areas, Brain Res. 177, 176-182

Murphy KM, Jones DG, Fenstemaker SB, Pegado VD, Kiorpes L, Movshon JA, 1998, Spacing of cytochrome oxidase blobs in visual cortex of normal and strabismic monkeys, Cereb. Cortex 8:237-44

Nicoll, R., and Jahr, C.E., 1982, Self-excitation of olfactory bulb neurones, Nature, 196:441-444

Nowycky, M.C., Mori, K. and Shepherd, G.M., 1981, GABAergic mechanisms of dendrodendritc synapses in isolated turtle olfactory bulb, J. Neurophysiol. 46, 639-648.

O'Keefe, J. and L. Nadel (1978). The Hippocampus as a Cognitive Map. Oxford, Clarendon Press.

Ottoson D., 1959, Studies of slow potentials in the rabbit's olfactory bulb and nasal mucosa, Acta. Physiol. Scand., 47:136-148.

Pavlides, C. and Winson, J. (1989) Influences of hippocampal place cell firing in the awake state on the activity of these cells during subsequent sleep episodes, J. Neuroscience 9:2907-2918.

Pinker, S., and Bloom, P., 1990, Natural language and natural selection. Behavioral and Brain Sciences 13:707-84.

Plaut, D. C. and Shallice, T. (1993): Deep dyslexia: A case study of connectionist neuropsychology. Cognitive Neuropsychology, 10:377-500.

Quillian, M.R., 1968, Semantic Memory, Semantic Information Processing (M.L. Minsky, Ed.), Cambridge, MA: The MIT Press, pp.216-270

Rizzolatti, G, and Arbib, M.A., 1998, Language Within Our Grasp, Trends in Neuroscience, 21(5):188-194.

Rizzolatti, G., Fadiga L., Gallese, V., and Fogassi, L. ,1996a, Premotor cortex and the recognition of motor actions. Cogn Brain Res., 3: 131-141.

Rizzolatti, G., Camarda, R., Fogassi, L., Gentilucci, M., Luppino, G. & Matelli, M. 1988 Functional Organization of Inferior Area 6 in the Macaque Monkey II. Area F5 and the Control of Distal Movements. Exp Brain Res. 71, 491-507

Rizzolatti, G., Fadiga, L., Matelli, M., Bettinardi, V., Perani, D., and Fazio, F., 1996, Localization of grasp representations in humans by positron emission tomography: 1. Observation versus execution. Exp Brain Res., 111:246-252.

Rozin, P., 1976, The evolution of intelligence and access to the cognitive unconscious, Prog. Psychobiol. Physiol. Psych., 6:245-280.

Rumelhart, D.E., and McClelland, J.L. (Eds.), 1986, Parallel Distributed Processing: Explorations in the Microstructure of Cognition, Cambridge, MA: A Bradford Book/The MIT Press.

Sakata, H., Taira, M., Murata, A. & Mine, S. 1992 Neural Mechanisms of Visual Guidance of Hand Action in the Parietal Cortex of the Monkey. Cerebral Cortex. 5, 429-38

Sherrington, C.S., 1906, The integrative action of the nervous system, New Haven and London, Yale University Press

Skarda, C.A., and Freeman, W.J., 1987, How brains make chaos in order to make sense of the world, Behav. Brain. Sci. 10:161-195.

Somogyi, P., 1977, A specific axonal interneuron in the visual cortex of the rat, Brain Res. 136:345-350.

Somogyi, P., Hodgson, A.J., and Smith, A.D., 1979, An approach to tracing neuron networks in the cerebral cortex and basal ganglia. Combination of Golgi staining, retrograde transport horseradish peroxidase and anterograde degeneration of synaptic boutons in the same material,

Somogyi, P. and Cowey, A., 1981, Combined Golgi and electron microscopic study on the synapses formed by double bouquet cells in the visual cortex of cat and monkey, J. Comp. Neurology, 195: 547-566.

Somogyi, P, Tamás, G, Lujan, R, Buhl, EH. 1998, Salient features of synaptic organization in the cerebral cortex. Brain. Res. Rev. 26:113-135.

Sperry, R,W. A modified concept of consciousness. Psychol. Rev. 76,195-206

Squire, LR and Zola SM, 1998, Episodic memory, semantic memory, and amnesia, Hippocampus 8:205-211.

Stern, CE and Hasselmo, ME., 1999, Bridging the gap; integrating cellular and functional magnetic resonance imaging studies of the hippocampus Hippocampus 9:45-53.

Swindale, N.V., 1990, Is the cerebral cortex modular?, Trends in Neurosciences, 13:487-492.

Swindale NV, 1998, Cortical organization: modules, polymaps and mosaics, Curr Biol. 8:R270-3

Szatmáry, Z, Lengyel, M, Érdi, P, Obermayer, K, (submitted): A Model of Place Field Formation Based on Recurring Consecutiveness of Perceptual Patterns.

Székely, Gy. and Czéh, G., 1971, Activity of spinal cord fragments and limbs deplanted in the dorsal fin of Urodele larvae, Acta Physiol. Acad. Sci. Hung. 40:303-312.

Szentágothai, J., 1967, The anatomy of complex integration units in the nervous system, in Recent development of neurobiology in Hungary I. Results in neuroanatomy, neuropharmacology and neurophysiology, (Lissák, K., ed.), Akadémiai Kiadó , Budapest, pp.9-45.

Szentágothai, J., 1969, Architecture of the cerebral cortex, Basic Mechanisms of the Epilepsies. (Jasper, H.H.I, Ward, A.A.Jr., and Pope, A. Eds.), Boston, Little Brown and Co., pp. 13-28.

Szentágothai, 1982, Too 'much' and too 'soon', Acta biol. Acad. Sci. Hung., 33:107-126.

Szentágothai, J., 1983 The modular architectonic principle of neural centers, Rev. Physiol. Biochem. Pharmacol. 98:11-61.

Szentágothai, J., 1984, Downward causation?, Ann.Rev. Neurosci., 7:1-11

Szentágothai, J., 1993, Self organization: the basic principle of neural functions, Theoretical Med. 14:101-116.

Szentágothai, J., and Arbib, M.A., 1975, Conceptual Models of Neural Organization, The MIT Press: Cambridge. MA. (Reprint of Szentágothai and Arbib 1974.)

Szentágothai, J., and Érdi, P., 1989, Self-organization in the nervous system, J. Soc. Biol. Struct. 12:367-384.

Taira, M., Mine, S., Georgopoulos, A. P., Murata, A. & Sakata, H. 1990 Parietal Cortex Neurons of the Monkey Related to the Visual Guidance of Hand Movement. Exp Brain Res. 83, 29-36

Traub, D., Wong, K.S. and Miles, R., 1987, In vitro models of epilepsy, in Neurotransmitters and Epilepsy, (Jobe PC and Laird HE II, eds.), The Humana Press, Clifton, NJ, pp. 161-190.

Traub, R.D., Miles, R., Muller, R.U. and Gulyás, A.I., 1992, Functional organization of the hippocampal CA3 region: implications for epilepsy, brain waves and spatial behavior, Network, 3: 465-488.

Traub, R.D. and Miles, R., 1991, Neuronal Networks of the Hippocampus, Cambridge Univ. Press 199

Traub, R.D. and Miles, R., 1992, Modeling hippocampal circuitry using data from whole cell patch clamp and dual intracellular recordings in vitro, Seminars in the Neurosciences 4:27-36.

Traub RD, Jefferys JG, Whittington MA, 1997 Simulation of gamma rhythms in networks of interneurons and pyramidal cells. J. Comput Neurosci., 4:141-50.

Tulving, E and Markowitsch, HJ, 1998, Episodic and declarative memory: role of the hippocampus. Hippocampus 8:198-204.

Ungerleider, L. G., and Mishkin, M., 1982, Two cortical visual systems, in Analysis of Visual Behavior (D. J. Ingle, M. A. Goodale, and R. J. W. Mansfield, Ed.), Cambridge, MA: The MIT Press

van Riemsdijk, H., and Williams, E. (1986): Introduction to the Theory of Grammar, Cambridge, MA: The MIT Press.

Vanier, M., and Caplan, D. (1990): CT scan correlates of agrammatism, in Agrammatic Aphasia, L. Menn and L. Obler, Eds., Amsterdam: Benjamins, pp.97-114.

Varga-Khadem, F, Gadian DG, Watkin KE, Conelly A, Van Paesschen W, Mishkin, M., 1997, Differential effects of early hippocampal pathology, on episodic and semantic memory. Science, 277:376-380.

Ventriglia, F., 1974, Kinetic approach to neural systems, I. Bull. Math. Biol. 36:534-544.

Ventriglia, F., 1994, Toward a kinetic theory of cortical-like neural fields, in Neural modeling and neural networks Ventriglia F. ed.), Pergamon Press, Oxford, pp. 217-249.

Wallenstein, G.V. and Hasselmo, M.E. ,1997, GABAergic modulation of hippocampal activity: Sequence learning, place field development, and the phase precession effect. Journal of Neurophysiology, 78: 393-408

Wang XJ, 1999, Fast burst firing and short-term synaptic plasticity: a model of neocortical chattering neurons. Neuroscience 89:347-62

Wang XJ and Buzsáki G, 1996, Gamma oscillation by synaptic inhibition in a hippocampal interneuronal network model. J. Neurosci. 16:6402-6413

Whittington, M.A., Traub, R.B. Jefferys, J., 1995, Synchronized oscillations in interneuron networks driven by metabotropic glutamate receptor activation, Nature 370: 612-615

Wilkins, K., and Wakefield, J., 1995, Brain Evolution and Neurolinguistic Preconditions, Behavioral and Brain Sciences.

Wilson, H.R. and Cowan, J., 1973, A mathematical theory of the functional dynamics of cortical and thalamic neurons tissue, Kybernetik l3:55-80

Záborszky, L., Palkovits, M. and Flerkó, B, 1992, A life-time adventure with the brain. An appreciation of his eigthies birthday, J. Comp. Neurol., 1992, 326:1-6.

Zhang, K., Ginzburg, I., McNaughton, B.L., Sejnowski, T.J., 1998, Interpreting neuronal population activity by reconstruction: Unified framework with application to hippocampal place cells. Journal of Neurophysiology, 79: 1017-1044

Zurif E (1984): Psycholinguistic Interpretations of the Aphasias. In: Biological Perspectives on Language, Caplan D, Lecours AR, Smith A, eds., Cambridge: MIT Press, pp 158-171.