Stabilized Best-of-K Training for Neural Combinatorial Optimization
Not on the map. Why?
Abstract (arXiv)
Leader Reward modifies POMO training to emphasize the best trajectory produced by repeated inference. We test a narrow extension: replace its binary leader/non-leader distinction with a stabilized rank signal indexed by a sampling budget K. With the POMO architecture, 3,050-epoch schedule, and TSP-100 test set held fixed, the Leader Reward reimplementation obtains 7.7662 under 100-start, 8-augmentation greedy decoding, matching the reported 7.766 at its displayed precision. Under independent sampling, the stabilized K=8 recipe lowers realized Best-of-8 cost in all three paired training seeds: 7.7944 versus 7.8136. This observation is estimation-only and decoder-specific: three seeds are below the six-seed testing floor, Leader Reward is better at sampled K=1, and it remains slightly better under its original augmented-greedy protocol. We make no unbiased-estimator, universal superiority, or state-of-the-art claim.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 4 of its 5 references that are in the map. First posted on arXiv 31 Jul 2026.
Builds on
4 of 5 references.
References · 4
- 2020POMO: Policy Optimization with Multiple Optima for Reinforcement Learningcited by 341
- 2025Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problemscited by 83
- 2022Winner Takes It All: Training Performant RL Populations for Combinatorial Optimizationcited by 57
- 2024Leader Reward for POMO-Based Neural Combinatorial Optimizationcited by 5
Built on by
Not cited yet.