pith. sign in

arxiv: 2309.06657 · v2 · pith:XMILIN6Cnew · submitted 2023-09-13 · 💻 cs.CL

Statistical Rejection Sampling Improves Preference Optimization

classification 💻 cs.CL
keywords policypreferenceslicoptimaloptimizationpairssamplinghuman
0
0 comments X
read the original abstract

Improving the alignment of language models with human preferences remains an active research challenge. Previous approaches have primarily utilized Reinforcement Learning from Human Feedback (RLHF) via online RL methods such as Proximal Policy Optimization (PPO). Recently, offline methods such as Sequence Likelihood Calibration (SLiC) and Direct Preference Optimization (DPO) have emerged as attractive alternatives, offering improvements in stability and scalability while maintaining competitive performance. SLiC refines its loss function using sequence pairs sampled from a supervised fine-tuned (SFT) policy, while DPO directly optimizes language models based on preference data, foregoing the need for a separate reward model. However, the maximum likelihood estimator (MLE) of the target optimal policy requires labeled preference pairs sampled from that policy. DPO's lack of a reward model constrains its ability to sample preference pairs from the optimal policy, and SLiC is restricted to sampling preference pairs only from the SFT policy. To address these limitations, we introduce a novel approach called Statistical Rejection Sampling Optimization (RSO) that aims to source preference data from the target optimal policy using rejection sampling, enabling a more accurate estimation of the optimal policy. We also propose a unified framework that enhances the loss functions used in both SLiC and DPO from a preference modeling standpoint. Through extensive experiments across three diverse tasks, we demonstrate that RSO consistently outperforms both SLiC and DPO on evaluations from both Large Language Model (LLM) and human raters.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 15 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Efficient Preference Poisoning Attack on Offline RLHF

    cs.LG 2026-05 unverdicted novelty 8.0

    Label-flip attacks on log-linear DPO reduce to binary sparse approximation problems that can be solved efficiently by lattice-based and binary matching pursuit methods with recovery guarantees.

  2. TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching

    cs.CL 2026-05 unverdicted novelty 7.0

    TBPO posits a token-level Bradley-Terry model and derives a Bregman-divergence density-ratio matching loss that generalizes DPO while preserving token-level optimality.

  3. TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching

    cs.CL 2026-05 unverdicted novelty 7.0

    Introduces TBPO, which derives a Bregman-divergence density-ratio matching objective for token-level preference optimization that generalizes DPO while preserving the induced optimal policy.

  4. Beyond Negative Rollouts: Positive-Only Policy Optimization with Implicit Negative Gradients

    cs.CL 2026-05 unverdicted novelty 7.0

    POPO uses bounded importance sampling on positive rollouts and a siamese policy network to achieve implicit negative gradients and stable optimization, matching or exceeding GRPO on math benchmarks such as 36.67% on A...

  5. Efficient Preference Poisoning Attack on Offline RLHF

    cs.LG 2026-05 unverdicted novelty 7.0

    Preference poisoning against log-linear DPO reduces to a binary sparse approximation problem solved by lattice-reduction (BAL-A) and matching-pursuit (BMP-A) algorithms that carry recovery guarantees.

  6. HiPO: Hierarchical Preference Optimization for Adaptive Reasoning in LLMs

    cs.AI 2026-04 unverdicted novelty 7.0

    HiPO improves LLM reasoning performance by optimizing preferences separately on response segments rather than entire outputs.

  7. Gradient-Guided Reward Optimization for Inference-time Alignment

    cs.CL 2026-06 unverdicted novelty 6.0

    GGRO monitors token entropy to trigger gradient-guided token injection from reward models, improving LLM alignment on safety, helpfulness, and reasoning tasks at inference time.

  8. TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching

    cs.CL 2026-05 unverdicted novelty 6.0

    TBPO derives a token-level preference optimization objective from sequence-level pairwise data via Bregman divergence ratio matching that generalizes DPO and improves alignment quality.

  9. Response Time Enhances Alignment with Heterogeneous Preferences

    cs.LG 2026-05 unverdicted novelty 6.0

    Response times modeled as drift-diffusion processes enable consistent estimation of population-average preferences from heterogeneous anonymous binary choices.

  10. Reasoning Structure Matters for Safety Alignment of Reasoning Models

    cs.AI 2026-04 unverdicted novelty 6.0

    Changing the internal reasoning structure of large reasoning models through simple supervised fine-tuning on 1K examples produces strong safety alignment that generalizes across tasks and languages.

  11. Beyond Importance Sampling: Rejection-Gated Policy Optimization

    cs.LG 2026-04 unverdicted novelty 6.0

    RGPO replaces importance sampling with a smooth [0,1] acceptance gate in policy gradients, unifying TRPO/PPO/REINFORCE, bounding variance for heavy-tailed ratios, and showing gains in online RLHF experiments.

  12. Enhancing the Reasoning Ability of Multimodal Large Language Models via Mixed Preference Optimization

    cs.CL 2024-11 conditional novelty 6.0

    Mixed Preference Optimization with the MMPR dataset boosts multimodal CoT reasoning, lifting InternVL2-8B to 67.0 accuracy on MathVista (+8.7 points) and matching the 76B model.

  13. SHE: Stepwise Hybrid Examination Reinforcement Learning Framework for E-commerce Search Relevance

    cs.AI 2025-10 unverdicted novelty 5.0

    SHE is a new RL framework using stepwise hybrid examination rewards to improve reasoning quality and accuracy in large-scale e-commerce query-product relevance prediction.

  14. BV-Blend: Uncertainty-Weighted Historical Baselines for Stable Critic-Free RL with Verifiable Rewards

    cs.AI 2026-06 unverdicted novelty 4.0

    BV-Blend blends prompt-local and semantic-cluster historical reward statistics via SEM-derived weights to stabilize critic-free RL advantage estimation.

  15. Supplement Generation Training for Enhancing Agentic Task Performance

    cs.LG 2026-04 unverdicted novelty 4.0

    SGT trains a lightweight model to generate task-specific supplemental text that improves performance of a larger frozen LLM on agentic tasks without modifying the large model.