Learning Admissible Heuristics via Cost Partitioning

In three linesNovel framework for learning admissible heuristics in optimal planning via cost partitioning. A neural network with axial self-attention predicts cost weights guaranteed admissible by construction, leveraging Lagrangian dual equivalence. Results show reduced node expansions versus suboptimal baselines while preserving optimality.

Summary generated by Claude — human-verified