JudgePanel: A Compact Judge with Panel Deliberation via Adaptive Multi-Reward Reinforcement Learning
This work proposes the first framework to equip a single compact judge with multi-agent panel deliberation capability at single-model inference cost, and introduces AdaReward, an adaptive multi-reward RL algorithm that dynamically rebalances reward component weights as different objectives saturate at different rates d...
Yi-Yue Qian, Shi-Nan Zhang, Huan Song et al.
· 0 citations