aiqbench analysis
This research introduces a theoretical framework for differentially private sparse Reinforcement Learning with Human Feedback (RLHF), addressing privacy risks in large language model alignment without suffering from prohibitive error bounds associated with dense parameterizations. By leveraging the assumption that human preferences rely on a small subset of relevant features, the proposed algorithms achieve superior privacy-utility trade-offs compared to existing dense baselines in sentiment and safety benchmarks. The work provides rigorous upper and lower bounds for both Sample-DP and Label-DP settings, offering a scalable solution for protecting sensitive user data during model training.
Differentially Private Sparse Reward Estimation with Preference Feedback
arXiv:2610.30099v1 Announce Type: new Abstract: Reinforcement Learning with Human Feedback (RLHF) has become a central paradigm for aligning Large Language Models with human values. However, the reliance on sensitive preference data necessitates rigorous differential privacy (DP) guarantees to prevent user information leakage. Existing private alignment methods typically assume dense reward parameterizations, resulting in estimation error bounds that scale polynomially with the ambient {model size $d$}. Such dependence is prohibitive in modern {overparameterized models} where $d$ is vast. In this work, we study \emph{Private Sparse RLHF} problem to address this limitation, assuming that human preferences (reward model) are governed by a subset of $s^* \ll d$ relevant features. We provide the first comprehensive theoretical framework for differentially private sparse reward estimation from pairwise preference feedback under the Bradley-Terry-Luce model, and introduce efficient algorithms for both Sample-DP and Label-DP settings. Theoretically, under the squared $\ell_2$ parameter estimation error, we establish upper bounds of $\widetilde{O}(s^*/n + {s^*}^2/n^2\varepsilon^2)$ for $(\varepsilon,\delta)$-Sample-DP and $\widetilde{O}(s^*/n + s^*/n\varepsilon^2)$ for $\varepsilon$-Label-DP based on randomized response, where $n$ is the training data size. For the lower bounds, we show that the Sample-DP rate is nearly optimal, and that the guarantee for our Label-DP method cannot be further improved within the randomized-response framework. We also provide a general lower bound for the problem with Label-DP. Empirically, our framework demonstrates superior efficacy over dense baselines across sentiment generation and safety alignment benchmarks, offering a significantly more favorable privacy-utility trade-off.
Abstract:Reinforcement Learning with Human Feedback (RLHF) has become a central paradigm for aligning Large Language Models with human values. However, the reliance on sensitive preference data necessitates rigorous differential privacy (DP) guarantees to prevent user information leakage. Existing private alignment methods typically assume dense reward parameterizations, resulting in estimation error bounds that scale polynomially with the ambient {model size $d$}. Such dependence is prohibitive in modern {overparameterized models} where $d$ is vast. In this work, we study \emph{Private Sparse RLHF} problem to address this limitation, assuming that human preferences (reward model) are governed by a subset of $s^* \ll d$ relevant features. We provide the first comprehensive theoretical framework for differentially private sparse reward estimation from pairwise preference feedback under the Bradley-Terry-Luce model, and introduce efficient algorithms for both Sample-DP and Label-DP settings. Theoretically, under the squared $\ell_2$ parameter estimation error, we establish upper bounds of $\widetilde{O}(s^*/n + {s^*}^2/n^2\varepsilon^2)$ for $(\varepsilon,\delta)$-Sample-DP and $\widetilde{O}(s^*/n + s^*/n\varepsilon^2)$ for $\varepsilon$-Label-DP based on randomized response, where $n$ is the training data size. For the lower bounds, we show that the Sample-DP rate is nearly optimal, and that the guarantee for our Label-DP method cannot be further improved within the randomized-response framework. We also provide a general lower bound for the problem with Label-DP. Empirically, our framework demonstrates superior efficacy over dense baselines across sentiment generation and safety alignment benchmarks, offering a significantly more favorable privacy-utility trade-off.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2610.30099 [cs.LG]
(or arXiv:2610.30099v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.30099 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 329 words
Related coverage
Learn-Then-Differentiate Gradient Estimation
arXiv · Oct 1, 2026
Computationally tractable robust differentially private mean estimation
arXiv · Aug 19, 2026
Differentially Private Nonparametric Modal Learning with Applications to Regression and Clustering
arXiv · Aug 3, 2026
Simultaneous Calibration of Noise Covariance and Kinematics for State Estimation of Legged Robots via Bi-level Optimization
arXiv · Jul 20, 2026