Research map / Learning theory and algorithms / Kernels and regression theory
Learning Admissible Heuristics for A*: Theory and Practice
Ehsan Futuhi, Nathan R. Sturtevant
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train with CEA loss instead of cross-entropy loss (ablation) | ResNet heuristic on 8-corner PDB · Rubik's Cube 8-corner PDB | overestimation rate: 2 x 10^-3 -> 3 x 10^-7 | Improved |
| train with CEA loss instead of cross-entropy loss (ablation) | ResNet heuristic on 6-edge PDB · Rubik's Cube 6-edge PDB | overestimation rate: 9 x 10^-2 -> 9 x 10^-5 | Improved |
| train with CEA loss instead of cross-entropy loss (ablation) | ResNet heuristic on 7-edge PDB · Rubik's Cube 7-edge PDB | average heuristic: 8.44 -> 7.45 | Hurt |
| use CEA-trained NN instead of size-matched min-compressed PDB | Learned heuristic · Rubik's Cube 8-corner PDB | average heuristic: 6.84 -> 8.76 | Improved |
| use CEA-trained NN instead of size-matched compressed PDB in search | A* with learned heuristic · Rubik's Cube 8-corner PDB | nodes generated: less than half | Improved |
| use CEA-trained NN instead of size-matched compressed PDB | Learned heuristic · Rubik's Cube 7-edge PDB | average heuristic: 6.83 -> 7.45 | Improved |
| use 8-bit quantized inference instead of 16-bit (ablation) | Learned heuristic · Rubik's Cube PDBs | accuracy | Hurt |
| use delta heuristic (subtract 4-edge PDB from 6-edge PDB) (ablation) | Learned heuristic · Rubik's Cube 6-edge PDB | class imbalance | Improved |
| increase number of training instances N (ablation) | Inadmissibility generalization error · Rubik's Cube 8-corner PDB | generalization error | Improved |
Closest papers by meaning
- Sample Complexity of Learning Heuristic Functions for Greedy-Best-First and A* Search (NeurIPS 2022 · similarity 0.87)
- A Unifying View on Individual Bounds and Heuristic Inaccuracies in Bidirectional Search (AAAI 2020 · similarity 0.84)
- Learning How to Cube (NeurIPS 2026 · similarity 0.84)
- Taking the Road Less Scheduled with Adaptive Polyak Steps (NeurIPS 2026 · similarity 0.82)
- Topological Generalization Bounds for Discrete-Time Stochastic Optimization Algorithms (NeurIPS 2024 · similarity 0.82)
- Generalization Bound of Gradient Flow through Training Trajectory and Data-dependent Kernel (NeurIPS 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
