Detection of Repetitive Patterns in Action Sequences with Noise in Programming by Demonstration

Abstract

Software applications that exploit implicit programming by demonstration should be able to detect repetitive patterns in user’s actions in an autonomous and efficient way. We present a software agent for the detection of repetitive action patterns that makes use of domain knowledge in this process. We explain its design rationale and discuss some of its advantages, by comparing it with the classic algorithm KRM, which does not make use of domain knowledge. We demonstrate that our agent might have a more efficient detection process for repetitive tasks since it activates the search algorithm fewer times. Moreover, we show that it can detect repetitive tasks even in the presence of noise in the action sequence.KeywordsState MachineDomain KnowledgeAction SequenceDetection ProcessGeneralizer AgentThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.