A Multi-Electrode Matrix for Studies of Temporal Signal Correlations Within Neural Assemblies

Abstract

The ability of biological sensory systems to recognize entities or patterns in the presence of background patterns, noise, masking, and in spite of an almost unlimited number of possible variations within a patterns class, greatly surpasses the performance of existing general-purpose computer-based pattern recognition systems. The human visual system, for example, enables us to recognize a complex pattern, such as a tree, as a well-defined entity — even if the tree is seen against a background of other trees. It is as if each part of the object had been “labeled” such that the whole pattern is perceived as an entity. A possible and efficient mechanism for such labeling, i. e. for the linking of regions and attributes that define the pattern, could be based on temporal signal correlations; this means that a certain fraction of the total number of action potentials in a given channel would be coherent (synchronous or having a constant phase lag) with part of the action potentials in other channels that belong to the same neural assembly encoding the pattern. A particular attraction of this scheme is that the pattern and its attributes can be easily separated from the background or from other patterns via an array of “synchronous switches”. An array of neurons whose threshold is modulated accordingly can thus act as a pattern filter¹.