A Cortical Approach Based on Cascaded Bidirectional Hidden Markov Models

Research in the field of neural processing proposes a bidirectional computation scheme among the hierarchical organized levels of the brain. This scheme is called cortical algorithm and can be realized using Cascaded Bidirectional Hidden Markov Models (CBHMMs). In this paper CBHMMs are investigated in the light of analysis-synthesis systems. Such systems are important elements of Cognitive Dynamic Systems and Cognitive User Interfaces. Some of the most salient properties of Cognitive Systems are their abilities to support inference and reasoning, planning under uncertainty and adaptation to changing environmental conditions. That is, beside the bidirectional computation scheme among the hierarchical organized levels, CBHMMs need to support logical operations like inference and reasoning. To integrate this new aspect to the analysis-synthesis framework we pick up an old suggestion from D.M. MacKay from the late 1960s. D.M. MacKay suggested to supplement Shannon’s measure of selective information content by a descriptive information content. Descriptive information in turn is composed of structural and metric information and considers the logical aspect of information.