A connectionist approach for learning search-control heuristics for automated deduction systems (Thesis, TU München, 1997) (Q2726241)

The booklet contributes to the field of automatic deduction. The central problem of automatic deduction is the explosive growth of search spaces when deduction length increases. The author presents a connectionist approach for learning search-control heuristics from examples of successful deductions. Representing and processing of symbolic structures (graphs, trees, terms, formulas etc.) is one of the most important topics in the connectionist community. In the current book the author presents a new and powerful approach for adaptive structure processing: folding architecture networks with the training method back-propagation through structure. The generality of these methods is demonstrated in experiments with a variety of classification problems.