Ebook: Models of Neural Networks II. Temporal Aspects of Coding and Information Processing in Biological Systems

Издательство Springer, 1995, -354 pp.The second volume of the Physics of Neural Networks series.Models of Neural Networks I (/file/1427677/)
Models of Neural Networks II. Temporal Aspects of Coding and Information Processing in Biological Systems (/file/1427678/)
Models of Neural Networks III. Association, Generalization, and Representation (/file/1427679/)
Models of Neural Networks IV. Early Vision and Attention (/file/1427681/)Since the appearance of Vol. 1 of Models of Neural Networks in 1991, the theory of neural nets has focused on two paradigms: information coding through coherent firing of the neurons and functional feedback.
Information coding through coherent neuronal firing exploits time as a cardinal degree of freedom. This capacity of a neural network rests on the fact that the neuronal action potential is a short, say 1 ms, spike, localized in space and time. Spatial as well as temporal correlations of activity may represent different states of a network. In particular, temporal correlations of activity may express that neurons process the same "object" of, for example, a visual scene by spiking at the very same time. The traditional description of a neural network through a firing rate, the famous S-shaped curve, presupposes a wide time window of, say, at least 100 ms. It thus fails to exploit the capacity to "bind" sets of coherently firing neurons for the purpose of both scene segmentation and figure-ground segregation.
Feedback is a dominant feature of the structural organization of the brain. Recurrent neural networks have been studied extensively in the physical literature, starting with the ground breaking work of John Hopfield (1982). Functional feedback arises between the specialized areas of the brain involved, for instance, in vision when a visual scene generates a picture on the retina, which is transmitted to the lateral geniculate body (LGN), the primary visual cortex, and then to areas with "higher" functions. This sequence looks like a feed-forward structure, but appearances are deceiving, for there are equally strong recurrent signals. One wonders what they are good for and how they influence or regulate coherent spiking. Their role is explained in various contributions to this volume, which provides an in-depth analysis of the two paradigms. The reader can enjoy a detailed discussion of salient features such as coherent oscillations and their detection, associative binding and segregation, Hebbian learning, and sensory computations in the visual and olfactory cortex.
Each volume of Models of Neural Networks begins with a longer paper that puts together some theoretical foundations. Here the introductory chapter, authored by Gerstner and van Hemmen, is devoted to coding and information processing in neural networks and concentrates on the fundamental notions that will be used, or treated, in the papers to follow.
More than 10 years ago Christoph von der Malsburg wrote the meanwhile classical paper "The correlation theory of brain function." For a long time this paper was available only as an internal report of the Max-Planck Institute for Biophysical Chemistry in Gottingen, Germany, and is here made available to a wide audience. The reader may verify that notions which seemed novel 10 years ago still are equally novel at present.
The paper "Firing rates and well-timed events in the cerebral cortex" by Moshe Abeles does exactly what its title announces. In particular, Abeles puts forward cogent arguments that the firing rate by itself does not suffice to describe neuronal firing. Wolf Singer presents a careful analysis of "The role of synchrony in neocortical processing and synaptic plasticity" and in so doing explains what coherent firing is good for. This essay is the more interesting since he focuses on the relation between coherence - or synchrony - and oscillatory behavior of spiking on a global, extensive scale.
This connection is taken up by Ritz et al. in their paper "Associative binding and segregation in a network of spiking neurons." Here one finds a synthesis of scene segmentation and binding in the associative sense of pattern completion in a network where neural coding is by spikes only. Moreover, a novel argument is presented to show that a hierarchical structure with feed-forward and feedback connections may play a dominant role in context sensitive binding. We consider this an explicit example of functional feedback as a "higher" area provides the context to data presented to several "lower" areas.
Coherent oscillations were known in the olfactory system long before they were discovered in the visual cortex. Zhaoping Li describes her work with John Hopfield in the paper "Modeling the sensory computations of the olfactory bulb." She shows that here too it is possible to describe both odor recognition and segmentation by the very same model.
Until now we have used the notions "coherence" and "oscillation" in a loose sense. One may ask: How can one attain the goal of "Detecting coherence in neuronal data?" Precisely this is explained by Klaus Pawelzik in his paper with the above title. He presents a powerful information-theoretic algorithm in detail and illustrates his arguments by analyzing real data. This is important not only for the experimentalist but also for the theoretician who wants to verify whether his model exhibits some kind of coherence and, if so, what kind of agreement with experiment is to be expected.
As is suggested by several papers in this volume, there seems to be a close connection between coherence and synaptic plasticity; see, for example, the essay by Singer {Secs. 13 and 14) and Chap. 1 by Gerstner and van Hemmen. Synaptic plasticity itself, a fascinating subject, is expounded by Brown and Chattarji in their paper "Hebbian synaptic plasticity." By now long-term depression is appreciated as an essential element of the learning process or, as Willshaw aptly phrased it, "What goes up must come down." On the other hand, Hebb's main idea, correlated activity of the preand postsynaptic neuron, has been shown to be a necessary condition for the induction of long-term potentiation but the appropriate time window of synchrony has not been determined unambiguously yet. A small time window in the millisecond range would allow to learn, store, and retrieve spatio-temporal spike patterns, as has been pointed out by Singer and implemented by the Hebbian algorithm of Gerstner et al. Whether or not such a small time window may exist is still to be shown experimentally.
A case study of functional feedback or, as they call it, reentry is provided by Sporns, Tononi, and Edelman in the essay "Reentry and dynamical interactions of cortical networks." Through a detailed numerical simulation these authors analyze the problem of how neural activity in the visual cortex is integrated given its functional organization in the different areas. In a sense, in this chapter the various parts of a large puzzle are put together and integrated so as to give a functional architecture. This integration, then, is sure to be the subject of a future volume of Models of Neural Networks.Coding and Information Processing in Neural Networks
The Correlation Theory of Brain Function
Firing Rates and Well-Timed Events in the Cerebral
The Role of Synchrony in Neocortical Processing and Synaptic Plasticity
Associative Binding and Segregation in a Network of Spiking Neurons
Modeling the Sensory Computations of the Olfactory Bulb
Detecting Coherence in Neuronal Data
Hebbian Synaptic Plasticity: Evolution of the Contemporary Concept
Reentry and Dynamical Interactions of Cortical Networks