rl
RL 训练的树枝该长在哪?别看哪里不确定,要看哪里的「梯度」最值钱
💡 一句话总结 :用树形 rollout 训练 LLM(在关键决策点分叉、探索多种走法)是 RL 训练的进阶玩法,但「树枝放在哪」一直靠直觉——通常是在模型不确定的地方多分支。Tzafon AI 的 EPIG-Tree 从数学上证明这不完整:正确的目标是「让策略梯度的估计方差最小」,由此推出两条分配律。多轮游戏胜率 0…
Computational tools for society’s most complex challenges
Associate Professor Cathy Wu uses reinforcement learning to help map out improvements to transportation and other multifaceted systems.
Computational tools for society’s most complex challenges
Associate Professor Cathy Wu uses reinforcement learning to help map out improvements to transportation and other multifaceted systems.
Capturing token IDs during agentic interactions for better reinforcement learning
A new Rust proxy called Turnstile sits between the model backend and the agent harness to capture information lost in mere text transcripts.






