On Neural Network Architecture Based on Concept Lattices

摘要

Selecting an appropriate network architecture is a crucial problem when looking for a solution based on a neural network. If the number of neurons in network is too high, then it is likely to overfit. Neural networks also suffer from poor interpretability of learning results. In this paper an approach to building neural networks based on concept lattices and on lattices coming from monotone Galois connections is proposed in attempt to overcome the mentioned difficulties.