CONTROL OF SENSORY PROCESSING- A HYPOTHESIS ON AND SIMULATION OF THE ARCHITECTURE OF AN ELEMENTARY CORTICAL PROCESSOR

The cortex with its high f lexibi l i ty of processing consists of a regular columnar structure, where the minicolumn as a f irmware structure of about 110 Neurons can be considered to be the elementary processing element. There is an interest ing model concerning this columnar structure based on localized groups of homogeneous formal elements with probabil ist ic features (2). However, those minicolumns are not a pool of randomly connected neurons but have a characterist ic layered structure and are composed of several distinct neuron types with specific processing characteristics. There is evidence, that these layers cannot be explained by a coding hierarchy (3) and only some part of that may be due to mapping processes between different phase spaces (4).What then is it for? In putting this problem into the framework of a definit ive Job in early visual processing, we submitted evidence that this layered structure represents a kind of f lexible control hierarchy which enables the manipulat ion of the thalamic interface such that the highly parallel sensory inputs to the cortex can be picked up as a sequence of lower dimensional parallel bytes, which are reassembled towards a oompletly parallel representat ion,in a way to ensure a rapid and smooth convergence of the interpretation process (I). We argue, that such a more complex structure of a single node in a network may be crucial for gett ing more f lexibi l i ty in tha network's behavior. 2. The model In Brain computing interpretation means the formation of an activity distr ibution i ~ the cortex which suff ic iently matches the activity distr ibution of sensory inputs at the interface to the cortex. For a couple of reasons (I), the thalamus can be regarded to be such an interface