VPD frames language feedback learning as variational EM so the teacher policy refines itself via trust-region updates on outcomes while the student learns dense token distributions on its own rollouts, outperforming fixed-teacher baselines on reasoning and code tasks.
hub Canonical reference
Direct preference optimization: Your language model is secretly a reward model
Canonical reference. 73% of citing Pith papers cite this work as background.
39
Pith papers citing it
Background
73% of classified citations
hub tools
citation-role summary
background 8
method 3
citation-polarity summary
representative citing papers
LLM simulators exhibit near-zero selective response to targeted misconception feedback and behave sycophantically, but SFT and SFS-aligned RL improve this property.
Self-distillation token rewards measure input-response-feedback pointwise mutual information, and CREDIT extracts the input-specific component with contrastive baselines to improve LLM reasoning performance.
ReCrit frames critic interaction as a correctness-transition problem and uses quadrant-based RL rewards to improve LLM performance on scientific reasoning benchmarks by rewarding corrections and robustness while penalizing sycophancy.
AgentForesight introduces an online auditor model that predicts decisive errors in multi-agent trajectories at the earliest step using a coarse-to-fine reinforcement learning recipe on a new curated dataset AFTraj-2K.
ArenaPO infers Gaussian capability distributions from pairwise preferences and applies truncated-normal latent inference to derive fine-grained offline rewards for preference optimization of text-to-image diffusion models.
ICU-Bench is a new continual unlearning benchmark for MLLMs using 1000 privacy profiles, 9500 images, and 100 forget tasks, showing existing methods fail to balance forgetting, utility, and scalability.
IRIS unifies self-play fine-tuning under an interpolative Rényi objective with adaptive alpha scheduling and reports better benchmark scores than baselines while surpassing full supervised fine-tuning with only 13% of the annotated data.
GenAC introduces generative critics with chain-of-thought reasoning and in-context conditioning to improve value approximation and downstream RL performance in LLMs compared to value-based and value-free baselines.
CLORE augments correct on-policy rollouts by deleting repetitive and irrelevant segments then optimizes with auxiliary DPO to improve accuracy-efficiency trade-off on math benchmarks.
A hierarchical variational formulation amortizes test-time guidance in diffusion models to achieve strong quality-speed tradeoffs with significantly reduced inference compute.
NFPO augments the PPO surrogate with N-step forward traces to bridge local approximations and exact policy gradients, delivering tighter policy-improvement bounds and improved results on reasoning benchmarks.
GPRL carries a k-dimensional skew-symmetric preference structure into policy updates with per-dimension advantages and a drift monitor, yielding 56.51% length-controlled win rate on AlpacaEval 2.0 from Llama-3-8B-Instruct while outperforming SimPO and SPPO on other benchmarks.
The paper introduces a dynamical model that decomposes alignment updates in LLM fine-tuning into rebound and driving forces and predicts a rehearsal priming effect.
SyncDPO improves temporal synchronization in video-audio joint generation using DPO with efficient on-the-fly negative sample construction and curriculum learning.
FATE lets LLM agents self-evolve safer behaviors by generating and filtering repairs from their own failure trajectories using verifiers and Pareto optimization.
MORA breaks the safety-helpfulness ceiling in LLMs by pre-sampling single-reward prompts and rewriting them to incorporate multi-dimensional intents, delivering 5-12.4% gains in sequential alignment and 4.6% overall improvement in simultaneous alignment.
Super-Linear Advantage Shaping (SLAS) introduces a non-linear geometric policy update for RL post-training of text-to-image models that reshapes the local policy space via advantage-dependent Fisher-Rao weighting to reduce reward hacking and improve performance over GRPO baselines.
G-Zero uses the Hint-δ intrinsic reward to drive co-evolution between a Proposer and Generator via GRPO and DPO, providing a theoretical suboptimality guarantee for self-improvement from internal dynamics alone.
DeltaRubric decomposes multimodal preference evaluation into self-generated planning and verification steps within a single model, producing large accuracy improvements on VL-RewardBench via multi-role reinforcement learning.
S-trace adds sparse eligibility traces to RLVR that mask low-entropy tokens, outperforming GRPO by 0.49-3.16% pass@16 on Qwen3 models while improving sample and token efficiency.
A contrastive visual forgetting technique constrained to the null space of retained knowledge enables targeted unlearning of visual concepts in MLLMs while preserving non-target visual and all textual knowledge.
RVPO penalizes variance across multiple reward signals during RLHF advantage aggregation, using a LogSumExp operator as a smooth variance penalty to reduce constraint neglect in LLM alignment.
HJA ranking separates consensus ranking, judge sensitivity, and residual disagreement as distinct inferential targets with identifiability conditions and an anchored alternating algorithm, yielding better recovery and uncertainty calibration than pooled baselines on synthetic and real data.
citing papers explorer
-
SyncDPO: Enhancing Temporal Synchronization in Video-Audio Joint Generation via Preference Learning
SyncDPO improves temporal synchronization in video-audio joint generation using DPO with efficient on-the-fly negative sample construction and curriculum learning.
-
Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion
MORA breaks the safety-helpfulness ceiling in LLMs by pre-sampling single-reward prompts and rewriting them to incorporate multi-dimensional intents, delivering 5-12.4% gains in sequential alignment and 4.6% overall improvement in simultaneous alignment.
-
G-Zero: Self-Play for Open-Ended Generation from Zero Data
G-Zero uses the Hint-δ intrinsic reward to drive co-evolution between a Proposer and Generator via GRPO and DPO, providing a theoretical suboptimality guarantee for self-improvement from internal dynamics alone.