Two new research papers introduce novel methods for learning admissible heuristics in AI planning and search. The first paper proposes a framework using Lagrangian dual equivalence and graph neural networks to infer cost partitions, guaranteeing admissibility. The second paper focuses on combinatorial search problems, developing a deep reinforcement learning approach with an underestimating Bellman operator and a post-hoc calibration to ensure heuristics never overestimate costs. AI
IMPACT These methods could improve the efficiency and optimality of AI planning and search algorithms for complex problems.
RANK_REASON Two academic papers published on arXiv presenting novel methods for learning admissible heuristics in AI.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →