Symbolic representation of neural networks

摘要: Neural networks often surpass decision trees in predicting pattern classifications, but their predictions cannot be explained. This algorithm's symbolic representations make each prediction explicit and understandable. Our approach to understanding a neural network uses rules represent the process. The algorithm, NeuroRule, extracts these from network. can interpreted by which, general, preserve accuracy explain We based NeuroRule on standard three layer feed forward consists of four phases. First, it builds weight decay backpropagation so that weights reflect importance network's connections. Second, prunes remove irrelevant connections units while maintaining predictive accuracy. Third, discretizes hidden unit activation values clustering. Finally, with discretized values.