Specific-to-general learning for temporal events with application to learning event definitions from video

摘要: We develop, analyze, and evaluate a novel, supervised, specific-to-general learner for simple temporal logic use the resulting algorithm to learn visual event definitions from video sequences. First, we introduce simple, propositional, temporal, event-description language called AMA that is sufficiently expressive represent many events yet restrictive support learning. then give algorithms, along with lower upper complexity bounds, subsumption generalization problems formulas. present positive-examples-only learning method based on these algorithms. also polynomial-time-computable "syntactic" test implies semantic without being equivalent it. A syntactic can be used in place of improve asymptotic algorithm. Finally, apply this task relational show it yields are competitive hand-coded ones.