Experiments on Adaptive Hashing Functions using Neural Networks

The generic function of a feedforward neural net 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 hashing function – the Neuro-Hasher. This paper reports the experiments on using a backpropagation network with a dynamic hidden layer for finding an appropriate hashing function for a given population of hashing keys. Comparative studies show that the Neuro-Hasher performs robustly over various populations of scarce hashing keys which would cause uneven distributions with some traditional hashing function.