Applications: Information Processing and Coding in Spatiotemporal Dynamical Systems

Abstract

In the previous chapter, we introduced a formalism for processing and coding information by a network of spiking neurons. The coding strategy assumed was a timing coding. One of the most interesting forms of timing coding is the synchronization of firing. In fact, several experimental works concentrate on the proof of such a form of coding (Gray and Singer, 1987, 1989; Eckhorn et al., 1988), especially on the relation to the question of binding of features. The binding problem refers to the mechanism of how the brain integrates information associated with different modules. Objects are generally believed to be represented by a collection of local features. For example, in the visual cortex it is well established that sensory neurons can be characterized by a receptive field that represents local visual features, such as edges, textures, colors, etc. The essential question therefore is how to link these local features that define an object. Classical paradigms for the integration of information in the brain can be embedded in either one of the two (opposite) theories of Hebb (1949) and Barlow (1972). Barlow postulates the existence of gnostic or grandmother cells, which are single neurons that integrate information yielding a local representation strategy. On the other hand, Hebb took the standpoint that a combination of multiple cells increasing their firing rates refers to linked features, assuming therefore a distributed representation strategy. The neurons that represent the local features of the object become active and constitute a so-called cell assembly. In the case of simultaneous activation of many cell assemblies, such as by the presence of several objects in visual images, it should be distinguished whether active cells belong to the same or to different cell assemblies.