Object-Feature Columns in the Monkey Inferior Temporal Cortex

A sattelite symposium at the European Neuroscience Meeting, "Computational processing in sensory systems" (Viena, Austria)1994.9.8
ヨーロッパ神経科学会サテライトシンポジウム 「感覚系における計算処理」


Ichiro Fujita
PREST, Research Development Corporation of Japan, and Laboratory for Neural Information Processing, Frontier Research Program, RIKEN, Wako, Saitama 351-O1, Japan.


Area TE of the inferior temporal cortex occupies the final stage of the monkey visual cortical pathway crucial for object recognition. TE cells do not respond to views of every object, but are activated by a subset of, often only a few, objects among hundreds of objects tested. These cells are selective for visual features shared by the objects, such as a particular shape, color, texture or a combination of these. Although stimulus features critical for activating TE cells are more complex than those at earlier stages of the visual cortices, they are still simpler than the complexity of a usual real-world object. Individual TE cells thus do not signal abstract representation of a particular object.

TE cells with similar selectivity cluster in multiple columns separated by those with different stimulus preferences. Stimuli which activate cells in one column do not excite cells in adjacent columns. Stimulus selectivity of columns is maintained stable at least over 30 days. The discreteness and temporal stability of TE columns provide us with a rationale to estimate the number of columns in area TE. The number is estimated to be 2000. The number of distinct object features represented in TE may be smaller than this number, because there are multiple columns selective for similar features.

When a monkey sees an object, a particular subset of columns, not the entire TE, nor a single column, nor a single "grandmother" cell, will be activated by the image. Different component features of the object activate different sets of columns. The composition of the entire image may be represented by the combination of active columns. One TE may hold only a few hundreds of components expressed in columns, but the principle of combination allows TE to represent an almost infinite number of objects. (end)