Implicit distributed associative memory for ill-posed object recognition

Neural nets are widely used in pattern recognition and classification problems. The number of patterns versus the number of characterizing features is an important factor affecting the performance of the net. If one has n input patterns each characterized by m features, a distributed associative memory (DAM) may be constructed to recall the closest pattern exemplar to an unknown pattern. If the input patterns are linearly dependent, several patterns may be recalled by the memory and the procedure is ill-posed. The author reformulates the DAM methodology as a linear inverse problem of estimating a response vector from a matrix of pattern exemplars and an unclassified input vector. Two techniques are discussed for estimating a response from this implicit DAM: singular value decomposition and a maximum entropy method. The influence of noise on the classification process is also investigated