2010/06/01 by Nan Li, Li, Nan, Subbarao Kambhampati +5
Computer Science · #AI-based Problem Solving and Planning #Artificial Intelligence (cs.AI) #Bayesian Modeling and Causal Inference #FOS: Computer and information sciences #Natural Language Processing Techniques #Semantic Web and Ontologies
paper · pdf · doi:10.48550/arxiv.1006.0274
openalex publication_date 2010/06/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We propose automatically learning probabilistic Hierarchical Task Networks\n(pHTNs) in order to capture a user's preferences on plans, by observing only\nthe user's behavior. HTNs are a common choice of representation for a variety\nof purposes in planning, including work on learning in planning. Our\ncontributions are (a) learning structure and (b) representing preferences. In\ncontrast, prior work employing HTNs considers learning method preconditions\n(instead of structure) and representing domain physics or search control\nknowledge (rather than preferences). Initially we will assume that the observed\ndistribution of plans is an accurate representation of user preference, and\nthen generalize to the situation where feasibility constraints frequently\nprevent the execution of preferred plans. In order to learn a distribution on\nplans we adapt an Expectation-Maximization (EM) technique from the discipline\nof (probabilistic) grammar induction, taking the perspective of task reductions\nas productions in a context-free grammar over primitive actions. To account for\nthe difference between the distributions of possible and preferred plans we\nsubsequently modify this core EM technique, in short, by rescaling its input.\n