Computational properties and behavioral expression of cortical-peripheral interactions suggested by a model of olfactory bulb and piriform cortex

A simulation of Layers I and II of olfactory (piriform) cortex, as connected to its primary input structure, olfactory bulb, and based on some of their most salient anatomical and physiological properties, is presented. The cortico-bulbar model produces a short sequence of distinct cortical responses on the presentation of a single simulated odorant, mediated by cortical-peripheral interactions during repetitive sampling at the theta rhythm. Its emergent computational properties, when coupled with synaptic long term potentiation, are studied and related to memory organization.;It is found that statistical properties of the training environment are reflected in the cortical encodings of input cues. In particular, clustering of similar cues is observed on each sampling episode, and the resultant sequences of cortical responses are found to reflect the hierarchical structure of the training environment. Examination of these sequences reveals that they correspond to a factorization of the input cue in terms of learning history.;Since olfactory cortex has direct and well defined projections to cortical and subcortical structures that play a prominent role in memory, its encoding properties are likely to be reflected in appropriate behavioral tasks. Experimental studies in the context of an odor discrimination paradigm prompted by predictions derived from the simulation are described, namely: sparse and stable encoding in piriform cortex; emergence of similarity based perceptual categories; and discrimination of components in a mixture contingent on experience. The results are found to be consistent with the model's predictions.;Simplification and analysis of the biological model led to identification of the basic operations that characterize its behavior, resulting in a novel, efficient class of algorithms for hierarchical unsupervised learning based on multi-sampling. Empirical results for one instantiation that performs hierarchical clustering are reported and shown to agree in general terms with those obtained with traditional methods.;Finally, the hypothesis is put forward that these cortico-bulbar networks and circuitry of similar design in other brain regions contain computational elements sufficient to construct perceptual hierarchies for use in recognizing environmental cues.