Neural Network Computations with Negative Triggering Thresholds

Abstract

Recent binary signal detection theoryand neural network assembly memory model’s optimal data-decoding/memory-retrieval algorithm exists simultaneously in functionally equivalent neural network (NN), convolutional, and Hamming distance forms. In present paper this NN algorithm has been specified to provide decoding/retrieval probabilities at both positive and negative neuron triggering thresholds needed, in particular, for ROC curve computations. Examples of intact and damaged NNs are considered, model neuron receptive fields are introduced, a comparison between NN and analytic computations of decoding/retrieval probabilities is also performed.