The self-organized map, an architecture suggested for artificial neural networks, is explained by presenting simulation experiments and practical applications. The self-organizing map has the property of effectively creating spatially organized internal representations of various features of input signals and their abstractions. One result of this is that the self-organization process can discover semantic relationships in sentences. Brain maps, semantic maps, and early work on competitive learning are reviewed. The self-organizing map algorithm (an algorithm which order responses spatially) is reviewed, focusing on best matching cell selection and adaptation of the weight vectors. Suggestions for applying the self-organizing map algorithm, demonstrations of the ordering process, and an example of hierarchical clustering of data are presented. Fine tuning the map by learning vector quantization is addressed. The use of self-organized maps in practical speech recognition and a simulation experiment on semantic mapping are discussed. >

The self-organizing map

In recent years, many new cortical areas have been identified in the macaque monkey. The number of identified connections between areas has increased even more dramatically. We report here on (1) a summary of the layout of cortical areas associated with vision and with other modalities, (2) a computerized database for storing and representing large amounts of information on connectivity patterns, and (3) the application of these data to the analysis of hierarchical organization of the cerebral cortex. Our analysis concentrates on the visual system, which includes 25 neocortical areas that are predominantly or exclusively visual in function, plus an additional 7 areas that we regard as visual-association areas on the basis of their extensive visual inputs. A total of 305 connections among these 32 visual and visual-association areas have been reported. This represents 31% of the possible number of pathways if each area were connected with all others. The actual degree of connectivity is likely to be closer to 40%. The great majority of pathways involve reciprocal connections between areas. There are also extensive connections with cortical areas outside the visual system proper, including the somatosensory cortex, as well as neocortical, transitional, and archicortical regions in the temporal and frontal lobes. In the somatosensory/motor system, there are 62 identified pathways linking 13 cortical areas, suggesting an overall connectivity of about 40%. Based on the laminar patterns of connections between areas, we propose a hierarchy of visual areas and of somatosensory/motor areas that is more comprehensive than those suggested in other recent studies. The current version of the visual hierarchy includes 10 levels of cortical processing. Altogether, it contains 14 levels if one includes the retina and lateral geniculate nucleus at the bottom as well as the entorhinal cortex and hippocampus at the top. Within this hierarchy, there are multiple, intertwined processing streams, which, at a low level, are related to the compartmental organization of areas V1 and V2 and, at a high level, are related to the distinction between processing centers in the temporal and parietal lobes. However, there are some pathways and relationships (about 10% of the total) whose descriptions do not fit cleanly into this hierarchical scheme for one reason or another. In most instances, though, it is unclear whether these represent genuine exceptions to a strict hierarchy rather than inaccuracies or uncertainities in the reported assignment.

/pdf/distributed-hierarchical-processing-in-the-primate-cerebral-1dvblm6d82.pdf

Distributed Hierarchical Processing in the Primate Cerebral Cortex

An overview of the self-organizing map algorithm, on which the papers in this issue are based, is presented in this article.

The Self-Organizing Map

It has been clear for almost two decades that cortical representations in adult animals are not fixed entities, but rather, are dynamic and are continuously modified by experience. The cortex can preferentially allocate area to represent the particular peripheral input sources that are proportionally most used. Alterations in cortical representations appear to underlie learning tasks dependent on the use of the behaviorally important peripheral inputs that they represent. The rules governing this cortical representational plasticity following manipulations of inputs, including learning, are increasingly well understood. In parallel with developments in the field of cortical map plasticity, studies of synaptic plasticity have characterized specific elementary forms of plasticity, including associative long-term potentiation and long-term depression of excitatory postsynaptic potentials. Investigators have made many important strides toward understanding the molecular underpinnings of these fundamental plasticity processes and toward defining the learning rules that govern their induction. The fields of cortical synaptic plasticity and cortical map plasticity have been implicitly linked by the hypothesis that synaptic plasticity underlies cortical map reorganization. Recent experimental and theoretical work has provided increasingly stronger support for this hypothesis. The goal of the current paper is to review the fields of both synaptic and cortical map plasticity with an emphasis on the work that attempts to unite both fields. A second objective is to highlight the gaps in our understanding of synaptic and cellular mechanisms underlying cortical representational plasticity.

/pdf/cortical-plasticity-from-synapses-to-maps-2j9eoi9428.pdf

CORTICAL PLASTICITY: From Synapses to Maps

The computational brain: Patricia S. Churchland and Terrence J. Sejnowski (MIT Press, Cambridge, MA, 1992); xi, 544 pages, $39.95

Microelectrode mapping experiments indicate that the classical primary somatosensory cortex of monkeys consists of as many as four separate body representations rather than just one. Two complete body surface representations occupy cortical fields 3b and 1. In addition, area 2 contains an orderly representation of predominantly "deep" body tissues. Area 3a may constitute a fourth representation.

https://science.sciencemag.org/content/sci/204/4392/521.full.pdf

Multiple representations of the body within the primary somatosensory cortex of primates.

Microelectrode multiunit mapping studies of parietal cortex in owl monkeys indicate that the classical "primary" somatosensory region (or "SI") including the separate architectonic fields 3a, 3b, 1, and 2 contains as many as four separate representations of the body rather than one, An analysis of receptive field locations for extensive arrays of closely placed recording sites in parietal cortex which were later related to cortical architecture led to the following conclusions: (1) There are two large systematic representations of the body surface within "SI." Each is activated by low threshold cutaneous stimuli; one representation is coextensive with Area 3b and the other with Area 1. (2) While each of these representations contains regions of cortex with topological or "somatotopic" transformations of skin surface, the representations have many discontinuities where adjoining skin surfaces are not adjoining in the representations. Thus, the representations can be considered as composites of somatotopically organized regions, but cannot be accurately depicted by simple continuous homunculi. Lines of discontinuity often cut across dermatomes and seldom follow dermatomal boundaries, i.e., neither cutaneous representation constitutes a systematic representation of dermatomal skin fields. (3) While the two cutaneous fields are basically similar in organization and are approxi- mate mirror images of each other, they differ in important details, i.e., lines of discontinuity in the representations and the sites of representations of dif- ferent specific skin surfaces differ significantly in the two representations. (4) The two cutaneous representations also differ in size and in the relative propor- tions of cortex devoted to representation of various body parts. Because the pro- portions in each representation differ, they cannot both be simple reflections of overall peripheral innervation density. (5) All or part of Area 2 contains a sys- tematic representation of deep body structures. These conclusions are consistent with a view of the anterior parietal region as containing functionally distinct fields at least partially related to different subsets of receptor populations and coding or representing different aspects of somatic sensation. We suggest that the "SI" region of primates be redefined as a parietal somatosensory strip, the Area 1 representation as the posterior cuta- neous field, and, for reasons of probable homology with "SI" of other mammals, the Area 3b representation as SI proper.

http://web.mit.edu/surlab/publications/1978_MerzenichKaasSurLin.pdf

Double representation of the body surface within cytoarchitectonic area 3b and 1 in “SI” in the owl monkey (aotus trivirgatus)

Subcortical projections of six visual cortical areas (Areas 17 and 18, the Middle Temporal, Dorsomedial and Medial Areas, and the Posterior Parietal Region) in the owl monkey, Aotus trivigatus, were investigated with autoradiographic methods following injections of tritiated proline. No contralateral projections were demonstrated. While some brainstem structures received input from all six subdivisions of cortex, each cortical area appeared to exhibit its own unique pattern of subcortical projections. All six cortical areas were found to project to the superior and inferior divisions of the pulvinar, reticular nucleus of the thalamus, pretectum and superior colliculus. Other subcortical targets of one or more visual cortical areas were the basal ganglia, claustrum, zona incerta, one or more of the intralaminar nuclei, lateral posterior nucleus, pregeniculate nucleus, dorsal lateral geniculate nucleus, and pontine nuclei. Furthermore, details of corticofugal projections to the dorsal lateral geniculate nucleus, pretectum and superior colliculus varied with the cortical area studied. The projections to the reticular nucleus, pregeniculate nucleus, dorsal lateral geniculate nucleus, the inferior and a portion of the superior division of the pulvinar and the superior colliculus were found to be topographically organized. The targets of the subcortical projections were compared with those of the retina, as revealed by autoradiographic methods following tritiated proline injections of the eye and were found to overlap to varying extents in the superior colliculus, pretectum and dorsal lateral geniculate nucleus and to be segregated in the pregeniculate nucleus. The results substantiate the validity of previous studies in the owl monkey that suggest that the visual cortex is subdivided into several functionally distinct areas; and illustrate the complexity of corticofugal influence on visual processing.

Subcortical projections of six visual cortical areas in the owl monkey, Aotus trivirgatus

Patterns of connections with other visual structures and architectonic characteristics were used to subdivide the inferior pulvinar complex of owl monkeys into three distinct nuclei termed the central inferior pulvinar, IPc, the medial inferior pulvinar, IPm, and the posterior inferior pulvinar, IPp. IPc occupies about 70%; IPm about 20%, and IPp about 10% of the inferior pulvinar complex. Encapsulating fiber bands distinguish the boundaries of the three nuceli. IPm is also identified by a much greater packing density of neurons than IPc and IPp. Both IPp and IPc receive input from the superior colliculus, but the terminations in IPp are denser. Visual cortical Areas 17, 18, MT, DM, M and PP (Allman and Kaas, '76) project to IPc and IPm in patterns that indicate that central vision is represented dorsorostrally and peripheral vision ventrocaudally in both nuclei. Terminations in IPm from Area MT are particularly dense. None of these visual areas projects to IPp. Rather, input to IPp appears to originate in cortex rostral to Area MT in the temporal lobe.

The inferior pulvinar complex in owl monkeys: architectonic subdivisions and patterns of input from the superior colliculus and subdivisions of visual cortex.

Efferent cortical projections of posterior parietal cortex were determined by degeneration and autoradiographic methods in owl monkeys. Intraregional connections were to the immediate surround of the injection or lesion site, and to distinct foci within the posterior parietal region. The extraregional ipsilateral connections were with (1) previously established subdivisions of visual association cortex (the Dorsomedial Area, the Medial Area, the Dorsolateral Area, and the Middle Temporal Area), (2) other locations in caudal neocortex, and (3) frontal cortex. The callosal projections were to separate foci in posterior parietal cortex of the contralateral cerebral hemisphere. The separate foci of both ipsilateral and contralateral terminations in posterior parietal cortex raise the possibility that this region contains more than one functional subdivision. The connections with visual association cortex suggest a role for parietal cortex in visual behavior. Other foci in caudal neocortex indicate the possible locations of additional subdivisions of association cortex.

C.-S. Lin

Papers

Multiple representations of the body within the primary somatosensory cortex of primates.

Double representation of the body surface within cytoarchitectonic area 3b and 1 in “SI” in the owl monkey (aotus trivirgatus)

Subcortical projections of six visual cortical areas in the owl monkey, Aotus trivirgatus

The inferior pulvinar complex in owl monkeys: architectonic subdivisions and patterns of input from the superior colliculus and subdivisions of visual cortex.

Cortical projections of posterior parietal cortex in owl monkeys