Group Robust Preference Optimization in Reward-free RLHF

• URL: http://arxiv.org/abs/2405.20304v1
• Date: Thu, 30 May 2024 17:50:04 GMT
• Title: Group Robust Preference Optimization in Reward-free RLHF
• Authors: Shyam Sundhar Ramesh, Yifan Hu, Iason Chaimalas, Viraj Mehta, Pier Giuseppe Sessa, Haitham Bou Ammar, Ilija Bogunovic,
• Abstract summary: We propose a novel Group Robust Preference Optimization (GRPO) method to align large language models to individual groups' preferences robustly. To achieve this, GRPO adaptively and sequentially weights the importance of different groups, prioritizing groups with worse cumulative loss. We significantly improved performance for the worst-performing groups, reduced loss imbalances across groups, and improved probability accuracies.
• Score: 23.622835830345725
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Adapting large language models (LLMs) for specific tasks usually involves fine-tuning through reinforcement learning with human feedback (RLHF) on preference data. While these data often come from diverse labelers' groups (e.g., different demographics, ethnicities, company teams, etc.), traditional RLHF approaches adopt a "one-size-fits-all" approach, i.e., they indiscriminately assume and optimize a single preference model, thus not being robust to unique characteristics and needs of the various groups. To address this limitation, we propose a novel Group Robust Preference Optimization (GRPO) method to align LLMs to individual groups' preferences robustly. Our approach builds upon reward-free direct preference optimization methods, but unlike previous approaches, it seeks a robust policy which maximizes the worst-case group performance. To achieve this, GRPO adaptively and sequentially weights the importance of different groups, prioritizing groups with worse cumulative loss. We theoretically study the feasibility of GRPO and analyze its convergence for the log-linear policy class. By fine-tuning LLMs with GRPO using diverse group-based global opinion data, we significantly improved performance for the worst-performing groups, reduced loss imbalances across groups, and improved probability accuracies compared to non-robust baselines.

Related papers

• Ordinal Preference Optimization: Aligning Human Preferences via NDCG [28.745322441961438]
  We develop an end-to-end preference optimization algorithm by approxing NDCG with a differentiable surrogate loss. OPO outperforms existing pairwise and listwise approaches on evaluation sets and general benchmarks like AlpacaEval.
  arXiv  Detail & Related papers  (2024-10-06T03:49:28Z)
• Pareto-Optimal Learning from Preferences with Hidden Context [18.340302968130683]
  We propose POPL, which frames discrepant group preferences as objectives with potential trade-offs. Our empirical evaluations demonstrate that POPL surpasses baseline methods in learning sets of reward functions. POPL can serve as a foundation for techniques optimizing specific notions of group fairness.
  arXiv  Detail & Related papers  (2024-06-21T18:57:38Z)
• Multi-Reference Preference Optimization for Large Language Models [56.84730239046117]
  We introduce a novel closed-form formulation for direct preference optimization using multiple reference models. The resulting algorithm, Multi-Reference Preference Optimization (MRPO), leverages broader prior knowledge from diverse reference models. Our experiments demonstrate that LLMs finetuned with MRPO generalize better in various preference data, regardless of data scarcity or abundance.
  arXiv  Detail & Related papers  (2024-05-26T00:29:04Z)
• MaxMin-RLHF: Towards Equitable Alignment of Large Language Models with Diverse Human Preferences [101.57443597426374]
  Reinforcement Learning from Human Feedback (RLHF) aligns language models to human preferences by employing a singular reward model derived from preference data. We learn a mixture of preference distributions via an expectation-maximization algorithm to better represent diverse human preferences. Our algorithm achieves an average improvement of more than 16% in win-rates over conventional RLHF algorithms.
  arXiv  Detail & Related papers  (2024-02-14T03:56:27Z)
• Improving Generalization of Alignment with Human Preferences through Group Invariant Learning [56.19242260613749]
  Reinforcement Learning from Human Feedback (RLHF) enables the generation of responses more aligned with human preferences. Previous work shows that Reinforcement Learning (RL) often exploits shortcuts to attain high rewards and overlooks challenging samples. We propose a novel approach that can learn a consistent policy via RL across various data groups or domains.
  arXiv  Detail & Related papers  (2023-10-18T13:54:15Z)
• Group Preference Optimization: Few-Shot Alignment of Large Language Models [28.464834028110538]
  Group Preference Optimization steers language models to preferences of individual groups in a few-shot manner. We empirically validate the efficacy of GPO through rigorous evaluations using large language models with varied sizes. Our results demonstrate that GPO not only aligns models more accurately but also requires fewer group-specific preferences, and less training and inference computing resources.
  arXiv  Detail & Related papers  (2023-10-17T18:41:57Z)
• Modeling the Q-Diversity in a Min-max Play Game for Robust Optimization [61.39201891894024]
  Group distributionally robust optimization (group DRO) can minimize the worst-case loss over pre-defined groups. We reformulate the group DRO framework by proposing Q-Diversity. Characterized by an interactive training mode, Q-Diversity relaxes the group identification from annotation into direct parameterization.
  arXiv  Detail & Related papers  (2023-05-20T07:02:27Z)
• Re-weighting Based Group Fairness Regularization via Classwise Robust Optimization [30.089819400033985]
  We propose a principled method, dubbed as ours, which unifies the two learning schemes by incorporating a well-justified group fairness metric into the training objective. We develop an iterative optimization algorithm that minimizes the resulting objective by automatically producing the correct re-weights for each group. Our experiments show that FairDRO is scalable and easily adaptable to diverse applications.
  arXiv  Detail & Related papers  (2023-03-01T12:00:37Z)
• Focus on the Common Good: Group Distributional Robustness Follows [47.62596240492509]
  This paper proposes a new and simple algorithm that explicitly encourages learning of features that are shared across various groups. While Group-DRO focuses on groups with worst regularized loss, focusing instead, on groups that enable better performance even on other groups, could lead to learning of shared/common features.
  arXiv  Detail & Related papers  (2021-10-06T09:47:41Z)
• Overcoming Data Sparsity in Group Recommendation [52.00998276970403]
  Group recommender systems should be able to accurately learn not only users' personal preferences but also preference aggregation strategy. In this paper, we take Bipartite Graphding Model (BGEM), the self-attention mechanism and Graph Convolutional Networks (GCNs) as basic building blocks to learn group and user representations in a unified way.
  arXiv  Detail & Related papers  (2020-10-02T07:11:19Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.