A NEURAL-NETWORK ASSOCIATIVE MEMORY FOR HANDWRITTEN CHARACTER-RECOGNITION USING MULTIPLE CHUA CHARACTERS

A neural network architecture and learning algorithm for associative m emory storage of analog patterns, continuous sequences, and chaotic at tractors in the same is described. System performance using many diffe rent attractors from the family of Chua attractors the Chua hardware c ircuit is investigated in an the problem of real time handwritten digi t recognition. of these attractors outperform the previously studied L orenz attractor system in terms of accuracy and speed of convergence I n the normal form projection algorithm, which was developed at Berkele y for associative memory storage of dynamic attractors, a matrix inver sion determines network weights, given prototype patterns to be stored . There are N units of capacity in an N node network with 3N2 weights. It costs one unit per static two per Fourier component of each period ic trajectory, and least three per chaotic attractor. There are no spu rious attractors and for periodic attractors there is a Liapunov funct ion a special coordinate system which governs the approach transient s tates to stored trajectories. Unsupervised or supervised incremental l earning algorithms for pattern classification, such as competitive lea rning or bootstrap Widrow-Hoff can easily be implemented. The architec ture can be ''folded'' into a recurrent network with higher order weig hts that can be used as a model of cortex that stores oscillatory and chaotic attractors by a Hebb rule. A novel computing architecture has been constructed of recurrently interconnected associative memory modu les of this Architectural variations employ selective synchronization of modules with chaotic attractors that communicate by broadspectrum c haotic signals to control the flow of computation.