1 paper · 1 filter
Shuai Lyu, Yuning Gong, Ruiling Gao +7
Group-based reinforcement learning methods for multimodal large language models typically rely on trajectory-level credit assignment that applies a single advantage to all tokens i…