A cortical sparse distributed coding model linking mini- and macrocolumn-scale functionality

摘要

No generic function for the minicolumn, i.e., one that would apply equallywell to all cortical areas and species, has yet been proposed. I propose thatthe minicolumn does have a generic functionality, which only becomes clear whenseen in the context of the function of the higher-level, subsuming unit, themacrocolumn. I propose that: a) a macrocolumn's function is to store sparsedistributed representations of its inputs and to be a recognizer of thoseinputs; and b) the generic function of the minicolumn is to enforcemacrocolumnar code sparseness. The minicolumn, defined here as a physicallylocalized pool of ~20 L2/3 pyramidals, does this by acting as a winner-take-all(WTA) competitive module, implying that macrocolumnar codes consist of ~70active L2/3 cells, assuming ~70 minicolumns per macrocolumn. I describe analgorithm for activating these codes during both learning and retrievals, whichcauses more similar inputs to map to more highly intersecting codes, a propertywhich yields ultra-fast (immediate, first-shot) storage and retrieval. Thealgorithm achieves this by adding an amount of randomness (noise) into the codeselection process, which is inversely proportional to an input's familiarity. Ipropose a possible mapping of the algorithm onto cortical circuitry, and adduceevidence for a neuromodulatory implementation of this familiarity-contingentnoise mechanism. The model is distinguished from other recent columnar corticalcircuit models in proposing a generic minicolumnar function in which a group ofcells within the minicolumn, the L2/3 pyramidals, compete (WTA) to be part ofthe sparse distributed macrocolumnar code.