Safe RLHF Beyond Expectation: Stochastic Dominance for Universal Spectral Risk Control

• URL: http://arxiv.org/abs/2603.10938v1
• Date: Wed, 11 Mar 2026 16:24:20 GMT
• Title: Safe RLHF Beyond Expectation: Stochastic Dominance for Universal Spectral Risk Control
• Abstract summary: We propose a risk-sensitive alignment framework that replaces expected cost constraints with First-Order Dominance (FSD) constraints.<n>We show that RAD improves harmlessness over baselines while remaining competitive in helpfulness, and exhibits greater robustness on out-of-distribution evaluations.
• Score: 12.053369001333058
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Safe Reinforcement Learning from Human Feedback (RLHF) typically enforces safety through expected cost constraints, but the expectation captures only a single statistic of the cost distribution and fails to account for distributional uncertainty, particularly under heavy tails or rare catastrophic events. This limitation is problematic when robustness and risk sensitivity are critical. Stochastic dominance offers a principled alternative by comparing entire cost distributions rather than just their averages, enabling direct control over tail risks and potential out-of-distribution failures that expectation-based constraints may overlook. In this work, we propose Risk-sensitive Alignment via Dominance (RAD), a novel alignment framework that replaces scalar expected cost constraints with First-Order Stochastic Dominance (FSD) constraints. We operationalize this constraint by comparing the target policy's cost distribution to that of a reference policy within an Optimal Transport (OT) framework, using entropic regularization and Sinkhorn iterations to obtain a differentiable and computationally efficient objective for stable end-to-end optimization. Furthermore, we introduce quantile-weighted FSD constraints and show that weighted FSD universally controls a broad class of Spectral Risk Measures (SRMs), so that improvements under weighted dominance imply guaranteed improvements in the corresponding spectral risk. This provides a principled mechanism for tuning a model's risk profile via the quantile weighting function. Empirical results demonstrate that RAD improves harmlessness over baselines while remaining competitive in helpfulness, and exhibits greater robustness on out-of-distribution harmlessness evaluations.

Related papers

• Capability Minimization as a Safety Primitive: Risk-Aware Causal Gating for Least-Privilege LLM Agents [0.0]
  Risk-Aware Causal Gating (RACG) is a framework that decides whether to act on, defer, or abstain from a model's prediction.<n>RACG models the causal pathway from candidate actions to outcomes and gates each decision according to an estimated counterfactual risk.<n>We show that RACG reduces high-cost errors substantially while preserving most of the utility of an ungated policy.
  arXiv  Detail & Related papers  (2026-06-11T20:19:35Z)
• Ratio-Variance Regularized Policy Optimization [64.95520246570446]
  We show that explicitly constraining the policy ratio variance provides a principled local approximation to trust-region constraints.<n>We introduce $bf R2bf VPO$ (Ratio-Variance Regularized Policy Optimization), which implements this constraint via a primal-dual optimization framework.
  arXiv  Detail & Related papers  (2026-05-26T09:53:42Z)
• DARC: Disagreement-Aware Alignment via Risk-Constrained Decoding [59.16244104797919]
  We propose **Disagreement-Aware Alignment via Risk-Constrained Decoding (DARC)**.<n>DARC frames response selection as distributionally robust, risk-sensitive decision making.<n>Experiments on alignment benchmarks show that DARC reduces disagreement and tail risk while maintaining competitive average quality.
  arXiv  Detail & Related papers  (2026-03-09T09:21:29Z)
• Conformal Thinking: Risk Control for Reasoning on a Compute Budget [60.65072883773352]
  Reasoning Large Language Models (LLMs) enable test-time scaling, with dataset-level accuracy improving as the token budget increases.<n>We re-frame the budget setting problem as risk control, limiting the error rate while minimizing compute.<n>Our framework introduces an upper threshold that stops reasoning when the model is confident and a novel lower threshold that preemptively stops unsolvable instances.
  arXiv  Detail & Related papers  (2026-02-03T18:17:22Z)
• Risk-Averse Constrained Reinforcement Learning with Optimized Certainty Equivalents [29.698100324454362]
  Constrained optimization provides a common framework for dealing with conflicting objectives in reinforcement learning (RL)<n>We propose a framework for risk-aware constrained RL, which exhibits per-stage properties jointly in reward values and time using optimized certainty equivalents (OCEs)<n>Our framework ensures an exact equivalent to the original constrained problem within a parameterized strong Lagrangian duality framework under appropriate constraint qualifications.
  arXiv  Detail & Related papers  (2025-10-23T04:33:32Z)
• Rectified Robust Policy Optimization for Model-Uncertain Constrained Reinforcement Learning without Strong Duality [53.525547349715595]
  We propose a novel primal-only algorithm called Rectified Robust Policy Optimization (RRPO)<n>RRPO operates directly on the primal problem without relying on dual formulations.<n>We show convergence to an approximately optimal feasible policy with complexity matching the best-known lower bound.
  arXiv  Detail & Related papers  (2025-08-24T16:59:38Z)
• COIN: Uncertainty-Guarding Selective Question Answering for Foundation Models with Provable Risk Guarantees [51.5976496056012]
  COIN is an uncertainty-guarding selection framework that calibrates statistically valid thresholds to filter a single generated answer per question.<n>COIN estimates the empirical error rate on a calibration set and applies confidence interval methods to establish a high-probability upper bound on the true error rate.<n>We demonstrate COIN's robustness in risk control, strong test-time power in retaining admissible answers, and predictive efficiency under limited calibration data.
  arXiv  Detail & Related papers  (2025-06-25T07:04:49Z)
• Data-Driven Calibration of Prediction Sets in Large Vision-Language Models Based on Inductive Conformal Prediction [0.0]
  We propose a model-agnostic uncertainty quantification method that integrates dynamic threshold calibration and cross-modal consistency verification.<n>We show that the framework achieves stable performance across varying calibration-to-test split ratios, underscoring its robustness for real-world deployment in healthcare, autonomous systems, and other safety-sensitive domains.<n>This work bridges the gap between theoretical reliability and practical applicability in multi-modal AI systems, offering a scalable solution for hallucination detection and uncertainty-aware decision-making.
  arXiv  Detail & Related papers  (2025-04-24T15:39:46Z)
• Conditional Conformal Risk Adaptation [9.559062601251464]
  We develop a new score function for creating adaptive prediction sets that significantly improve conditional risk control for segmentation tasks.<n>We introduce a specialized probability calibration framework that enhances the reliability of pixel-wise inclusion estimates.<n>Our experiments on polyp segmentation demonstrate that all three methods provide valid marginal risk control and deliver more consistent conditional risk control.
  arXiv  Detail & Related papers  (2025-04-10T10:01:06Z)
• Model-Based Epistemic Variance of Values for Risk-Aware Policy Optimization [59.758009422067]
  We consider the problem of quantifying uncertainty over expected cumulative rewards in model-based reinforcement learning. We propose a new uncertainty Bellman equation (UBE) whose solution converges to the true posterior variance over values. We introduce a general-purpose policy optimization algorithm, Q-Uncertainty Soft Actor-Critic (QU-SAC) that can be applied for either risk-seeking or risk-averse policy optimization.
  arXiv  Detail & Related papers  (2023-12-07T15:55:58Z)
• f-FERM: A Scalable Framework for Robust Fair Empirical Risk Minimization [9.591164070876689]
  This paper presents a unified optimization framework for fair empirical risk based on f-divergence measures (f-FERM) In addition, our experiments demonstrate the superiority of fairness-accuracy tradeoffs offered by f-FERM for almost all batch sizes. Our extension is based on a distributionally robust optimization reformulation of f-FERM objective under $L_p$ norms as uncertainty sets.
  arXiv  Detail & Related papers  (2023-12-06T03:14:16Z)

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.