Invariant object recognition in the visual system with error correction and temporal difference learning

it has been proposed that invariant pattern recognition might be implemente d using a learning rule that utilizes a trace of previous neural activity w hich, given the spatio-temporal continuity of the statistics of sensory inp ut, is likely to be about the same object though with differing transforms in the short time scale. Recently, it has been demonstrated that a modified Hebbian rule which incorporates a trace of previous activity but no contri bution from the current activity can offer substantially improved performan ce. In this paper we show how this rule can be related to error correction rules, and explore a number of error correction rules that can be applied t o and can produce good invariant pattern recognition. An explicit relations hip to temporal difference learning is then demonstrated, and from this fur ther learning rules related to temporal difference learning are developed. This relationship to temporal difference learning allows us to begin to exp loit established analyses of temporal difference learning to provide a theo retical framework for better understanding the operation and convergence pr operties of these learning rules, and more generally, of rules useful for l earning invariant representations. The efficacy of these different rules fo r invariant object recognition is compared using VisNet, a hierarchical com petitive network model of the operation of the visual System.