Randomness as a constraint
Some problems require solutions that “look random”, with no apparent patterns and perhaps passing statistical tests of randomness. The reason could be to avoid statistical bias, to avoid detection, or to behave unpredictably. Based on ideas from data compression and algorithmic information theory, we propose “entropy constraint” to allow only solutions with few or no discernible patterns. Entropy constraints can be implemented in Constraint Programming using well-known global constraints. We apply them to a real problem from experimental psychology, and to an artificial factory inspection problem. This work was done with Armagan Tarim and Roberto Rossi, and follows our ECAI'14 paper on "Statistical Constraints".
Lecturer in UCC Insight
Tuesday, 25 November, 2014 (All day)