A Pattern Recognition Model based on Invariant Representations of Space Time Data

摘要

This paper proposes a generic pattern recognition model based on invariant representations of space time data, this model is mainly based on known operation aspects of the neocortex and it is oriented to deal with sequences of patterns. The theoretical model also establishes a more general framework for treatment of space time data through a dimensionality reduction process. For a given instance of space time data, the process characterizes a space time region that might be called an invariant representation. The model exhibits desirable properties for a pattern recognition system, such as spatial and temporal autoassociativity, spatial and temporal noise tolerance, recognition under sequence contextualization, and input prediction.