arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2401.04056cs.LG
收藏

一种极小极大主义的人类反馈强化学习方法

A Minimaximalist Approach to Reinforcement Learning from Human Feedback

Carnegie Mellon University(卡内基梅隆大学);Google Research(谷歌研究院)

Gokul Swamy, Christoph Dann, Rahul Kidambi, Zhiwei Steven Wu, Alekh Agarwal

AI总结:

本文提出自博弈偏好优化(SPO),将人类偏好学习构建为零和博弈,通过单智能体自对弈避免奖励模型训练和对抗训练,可证明处理非马尔可夫、不可传递及随机偏好,并在连续控制任务上比基于奖励模型的方法更高效且更鲁棒。

摘要:

我们提出自博弈偏好优化(SPO),一种用于从人类反馈中进行强化学习的算法。我们的方法具有极简主义特征,因为它不需要训练奖励模型,也不需要不稳定的对抗训练,因此实现起来相当简单。我们的方法又具有极大主义特征,因为它可证明地处理非马尔可夫、不可传递和随机偏好,同时对困扰离线序列预测方法的复合误差具有鲁棒性。为了实现上述性质,我们基于极小极大赢家(MW)这一概念进行构建,这是社会选择理论文献中的一种偏好聚合概念,它将从偏好中学习框架化为两个策略之间的零和博弈。通过利用该博弈的对称性,我们证明,与其使用传统的两个策略对决技术来计算MW,我们可以简单地让单个智能体与自身对弈,同时保持强收敛保证。在实践中,这对应于从一个策略中采样多条轨迹,让一个偏好模型或教师模型对它们进行比较,然后使用获胜比例作为特定轨迹的奖励。我们证明,在一组连续控制任务上,我们能够比基于奖励模型的方法学习效率显著更高,同时在聚合人类判断时实践中经常出现的不可传递和随机偏好下保持鲁棒性。

英文摘要:

We present Self-Play Preference Optimization (SPO), an algorithm for reinforcement learning from human feedback. Our approach is minimalist in that it does not require training a reward model nor unstable adversarial training and is therefore rather simple to implement. Our approach is maximalist in that it provably handles non-Markovian, intransitive, and stochastic preferences while being robust to the compounding errors that plague offline approaches to sequential prediction. To achieve the preceding qualities, we build upon the concept of a Minimax Winner (MW), a notion of preference aggregation from the social choice theory literature that frames learning from preferences as a zero-sum game between two policies. By leveraging the symmetry of this game, we prove that rather than using the traditional technique of dueling two policies to compute the MW, we can simply have a single agent play against itself while maintaining strong convergence guarantees. Practically, this corresponds to sampling multiple trajectories from a policy, asking a preference or teacher model to compare them, and then using the proportion of wins as the reward for a particular trajectory. We demonstrate that on a suite of continuous control tasks, we are able to learn significantly more efficiently than reward-model based approaches while maintaining robustness to the intransitive and stochastic preferences that frequently occur in practice when aggregating human judgments.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑