The Power of Active Multi-Task Learning in Reinforcement Learning from Human Feedback
This work addresses the reliance on large human-labeled data in RLHF, which is a bottleneck for improving large language models, by proposing an incremental method to reduce sample complexity through active multi-task learning.
The paper tackles the high sample complexity of reinforcement learning from human feedback (RLHF) by formulating it as a contextual dueling bandit problem with a common linear representation, and shows that considering task relevance can significantly reduce the sample complexity of source tasks compared to uniform sampling, with the target task complexity linear in the latent dimension.
Reinforcement learning from human feedback (RLHF) has contributed to performance improvements in large language models. To tackle its reliance on substantial amounts of human-labeled data, a successful approach is multi-task representation learning, which involves learning a high-quality, low-dimensional representation from a wide range of source tasks. In this paper, we formulate RLHF as the contextual dueling bandit problem and assume a common linear representation. We demonstrate that the sample complexity of source tasks in multi-task RLHF can be reduced by considering task relevance and allocating different sample sizes to source tasks with varying task relevance. We further propose an algorithm to estimate task relevance by a small number of additional data and then learn a policy. We prove that to achieve $\varepsilon-$optimal, the sample complexity of the source tasks can be significantly reduced compared to uniform sampling. Additionally, the sample complexity of the target task is only linear in the dimension of the latent space, thanks to representation learning.