Abstract
More than 50 years ago Bongard introduced 100 visual concept learningproblems as a testbed for intelligent vision systems. These problems are nowknown as Bongard problems. Although they are well known in the cognitivescience and AI communities only moderate progress has been made towardsbuilding systems that can solve a substantial subset of them. In the systempresented here, visual features are extracted through image processing and thentranslated into a symbolic visual vocabulary. We introduce a formal languagethat allows representing complex visual concepts based on this vocabulary.Using this language and Bayesian inference, complex visual concepts can beinduced from the examples that are provided in each Bongard problem. Contraryto other concept learning problems the examples from which concepts are inducedare not random in Bongard problems, instead they are carefully chosen tocommunicate the concept, hence requiring pragmatic reasoning. Taking pragmaticreasoning into account we find good agreement between the concepts with highposterior probability and the solutions formulated by Bongard himself. Whilethis approach is far from solving all Bongard problems, it solves the biggestfraction yet.