REVIEW 13 cited by
Statistical Rejection Sampling Improves Preference Optimization
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
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.
Forward citations
Cited by 13 Pith papers
-
Test-Time Scaling via Error Localization
TTEL uses feedback-induced token probability drops to localize the first error in a failed reasoning trace and branch a new generation from that prefix, improving pass@k per token on coding and math benchmarks.
-
Multi-Turn On-Policy Distillation with Prefix Replay
ReOPD offline-distills multi-turn agentic LLMs via teacher-prefix replay plus step-decay sampling, matching online OPD accuracy at ≥4× speed with zero tool calls.
-
Do LLM Modules Generalize? A Study on Motion Generation for Autonomous Driving
On Waymo Sim Agents, LLM-style tokenization, positional embeddings, pretraining, RL post-training, and test-time search can be adapted to improve motion generation, but not all transfer without domain-specific changes.
-
Bridging Offline and Online Reinforcement Learning for LLMs
Semi-online DPO, which syncs the generation model every few update steps, performs nearly as well as fully online DPO and GRPO, while strongly beating offline DPO.
-
ReCUT: Balancing Reasoning Length and Accuracy in LLMs via Stepwise Trails and Preference Optimization
ReCUT combines stepwise long-short sampling, dual DPO training, and DARE-Ties parameter interpolation to compress LLM reasoning chains by 30-50% without sacrificing accuracy on math benchmarks.
-
Thompson Sampling in Online RLHF with General Function Approximation
A model-free posterior sampling algorithm for online RLHF is shown to achieve O(sqrt(T)) regret when the completed function class has low Bellman eluder dimension.
-
Square$\chi$PO: Differentially Private and Robust $\chi^2$-Preference Optimization in Offline Direct Alignment
SquareχPO, a square-loss variant of χPO, achieves optimal 1/sqrt(n) suboptimality under label privacy and Huber corruption for offline direct alignment with general function classes.
-
Learning a Pessimistic Reward Model in RLHF
Pessimistic fine-tuning of reward models against rejection-sampling policies lets RLHF agents optimize greedily without KL regularization and still avoid reward hacking.
-
Temporal Self-Rewarding Language Models: Decoupling Chosen-Rejected via Past-Future
Anchoring rejected responses to the initial model and choosing responses from a future model raises AlpacaEval 2.0 win rate from 19.69 to 29.44 for Llama3.1-8B.
-
Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences
SmPO-Diffusion improves diffusion-model preference alignment with reward-model soft labels and ReNoise inversion, reporting higher human-preference scores and up to 26x lower training cost than Diffusion-KTO.
-
Modeling and Optimizing User Preferences in AI Copilots: A Comprehensive Survey and Taxonomy
AI copilot preference optimization is organized into a pre-, mid-, and post-interaction taxonomy, with a unified definition of AI copilots.
-
PKG-DPO: Optimizing Domain-Specific AI systems with Physics Knowledge Graphs and Direct Preference Optimization
A physics knowledge graph plus Direct Preference Optimization reduces physically invalid AI recommendations in welding by 17% over knowledge-graph-only DPO, but the evidence is weakened by circular evaluation and no p...
-
Inverse Reinforcement Learning Meets Large Language Model Post-Training: Basics, Advances, and Opportunities
A tutorial reviewing LLM alignment through the lens of inverse reinforcement learning, arguing that neural reward models learned from human data are central to post-training.
Discussion (0). Sign in to comment.