How to cluster in parallel with neural networks

It is shown that the problem of partitioning a net of N patterns in a d-dimensional metric space into K clusters can, in spite of its exponential complexity, be formulated as an optimization problem for which very good, but not necessary optimal, solutions can be found by using a neural network. The network starts from many randomly selected initial states. The network is simulated on the MPP (a 128*128 SIMD array machine), where the massive parallelism is used not only in solving the differential equations that govern the evolution of the network, but also by starting the network from many initial states at once, thus obtaining many solutions in one run. Speedups of two to three orders of magnitude over serial implementations are obtained. Through analog VLSI implementations, speedups commensurate with human perceptual abilities should be possible.<
>