Minimal Perceptrons for Memorizing Complex Patterns
This work addresses the fundamental challenge of designing optimal neural network architectures for specific tasks, which is incremental as it builds on existing knowledge of pattern memorization.
The study tackled the problem of determining the minimal network size needed for three-layered neural networks to memorize binary patterns, by developing a new complexity measure and predicting minimal sizes for regular, random, and complex patterns, with predictions validated through simulations using back-propagation.
Feedforward neural networks have been investigated to understand learning and memory, as well as applied to numerous practical problems in pattern classification. It is a rule of thumb that more complex tasks require larger networks. However, the design of optimal network architectures for specific tasks is still an unsolved fundamental problem. In this study, we consider three-layered neural networks for memorizing binary patterns. We developed a new complexity measure of binary patterns, and estimated the minimal network size for memorizing them as a function of their complexity. We formulated the minimal network size for regular, random, and complex patterns. In particular, the minimal size for complex patterns, which are neither ordered nor disordered, was predicted by measuring their Hamming distances from known ordered patterns. Our predictions agreed with simulations based on the back-propagation algorithm.