Reinforcement Learning from Human Feedback (RLHF) has emerged as a pivotal technique for aligning artificial intelligence systems with human values, achieving remarkable success in fine-tuning large language models. However, existing RLHF frameworks often assume that human preferences are relatively homogeneous and can be captured by a single, unified reward model. This assumption overlooks the inherent diversity and heterogeneity across individuals, limiting the adaptability of RLHF to personalized scenarios and risking misalignments that can diminish user satisfaction and trust in AI systems. In this paper, we address these challenges by introducing Low-Rank Adaptation (LoRA) into the personalized RLHF framework. We apply LoRA in the the aggregated parameter space of all personalized reward functions, thereby enabling efficient learning of personalized reward models from potentially limited local datasets. Our approach exploits potential shared structures among the local ground-truth reward models while allowing for individual adaptation, without relying on restrictive assumptions about shared representations as in prior works. We further establish sample complexity guarantees for our method. Theoretical analysis demonstrates the effectiveness of the proposed approach in capturing both shared and individual-specific structures within heterogeneous human preferences, addressing the dual challenge of personalization requirements and practical data constraints. Experimental results on real-world datasets corroborate the efficiency of our algorithm in the personalized RLHF setting.

An intriguing property of the Transformer is its ability to perform in-context learning (ICL), where the Transformer can solve different inference tasks without parameter updating based on the contextual information provided by the corresponding input-output demonstration pairs. It has been theoretically proved that ICL is enabled by the capability of Transformers to perform gradient-descent algorithms (Von Oswald et al., 2023a; Bai et al., 2024). This work takes a step further and shows that Transformers can perform learning-to-optimize (L2O) algorithms. Specifically, for the ICL sparse recovery (formulated as LASSO) tasks, we show that a K-layer Transformer can perform an L2O algorithm with a provable convergence rate linear in K. This provides a new perspective explaining the superior ICL capability of Transformers, even with only a few layers, which cannot be achieved by the standard gradient-descent algorithms. Moreover, unlike the conventional L2O algorithms that require the measurement matrix involved in training to match that in testing, the trained Transformer is able to solve sparse recovery problems generated with different measurement matrices. Besides, Transformers as an L2O algorithm can leverage structural information embedded in the training tasks to accelerate its convergence during ICL, and generalize across different lengths of demonstration pairs, where conventional L2O algorithms typically struggle or fail. Such theoretical findings are supported by our experimental results.

In the development of machine learning (ML), the role of representation learning has become increasingly essential. It transforms raw data into meaningful features, reveals hidden patterns and insights in data, and facilitates efficient learning of various ML tasks such as meta-learning (Tripuraneni et al., 2021), multi-task learning (Wang et al., 2016a), and few-shot learning (Du et al., 2020). Recently, representation learning has been introduced to the federated learning (FL) framework to cope with the heterogeneous local datasets at participating clients (Liang et al., 2020). In the FL setting, it often assumes that all clients share a common representation, which works in conjunction with personalized local heads to realize personalized prediction while harnessing the collective training power (Arivazhagan et al., 2019; Collins et al., 2021; Zhong et al., 2022; Shen et al., 2023). Existing theoretical analysis of representation learning usually assumes the adopted model is overparameterized to almost fit the ground-truth model (Tripuraneni et al., 2021; Wang et al., 2016a). While this may be valid for expressive models like Deep Neural Networks (He et al., 2016; Liu et al., 2017) or Large Language Models (OpenAI, 2023; Touvron et al., 2023), it may be too restrictive for FL on resource-constrained devices, as adopting over-parameterized models in such a framework faces several significant challenges, as elaborated below.