Experiments on estimating random mapping

The generic function of a feedforward multilayer perceptron (MLP) network is to map patterns from one space to another. This mapping function, determined by the set of examples used to train the network, may be viewed as a hash function. This paper reports the experiments on using a backpropagation MLP network with a dynamic hidden layer to estimate a random mapping from the input to output space and use this estimated mapping as a hash function for a given population of keys. Comparative studies show that the MLP estimated hash functions performs robustly over various population of scarce hash keys which would cause uneven distributions with some traditional hash function