Abstract: Many HEP data analysis applications can be formulated as the pattern recognition or classification task. Behind the standard statistical methods the only widespread and powerful technique in use seems to be
the feed forward NN classifier. One can argue, however, that as the network is a black-box system, an unpleasant consequence of the refinement of the knowledge by a neural net training is that the user loses the interpretation of the results. Opposite to NN, the knowledge by the rule induction algorithms (RIA) is much more understandable and allows further "common sense" tuning. This paper provides the first results of my experiments with simple RIA's based on the attribute valued logic applied to HEP data analysis. The emphasise is given on the evaluation of the different search and pruning heuristics.I took several challenging tasks from the LHC physics as a benchmark.

Back to Program

Back to ACAT Web page