Anchor-guided variance-aware reward modeling uses two response-level anchors to resolve non-identifiability in Gaussian models of pluralistic preferences, yielding provable identification, a joint training objective, and improved RLHF performance.
hub
arXiv preprint arXiv:2502.18770 , year=
13 Pith papers cite this work. Polarity classification is still indexing.
13
Pith papers citing it
hub tools
citation-role summary
background 3
citation-polarity summary
roles
background 3polarities
background 3representative citing papers
LPA uses fewer than 100 personality trait statements to train LLMs for harmlessness, matching the robustness of methods using 150k+ harmful examples while generalizing better to new attacks.
Introduces the first benchmark for fine-grained failures in reinforcement fine-tuning of LLMs and an automatic management framework that detects, diagnoses, and remediates them.
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.
Training-inference mismatch in separated rollout and optimization stages of LLM RL can independently cause training collapse.
Rubric-based RL verifiers can be gamed via partial criterion satisfaction and implicit-to-explicit tricks, yielding proxy gains that do not improve quality under rubric-free judges; stronger verifiers reduce but do not eliminate the mismatch.
SelectiveRM applies optimal transport with a joint consistency discrepancy and partial mass relaxation to produce reward models that optimize a tighter upper bound on clean risk while autonomously dropping noisy preference samples.
A factored causal representation learning method improves robustness of reward models in RLHF by isolating causal factors from biases like length and sycophancy using adversarial gradient reversal.
Process supervision via RAG-Gym produces more reliable and generalizable search agents, with gains driven by higher-quality queries on out-of-domain multi-hop tasks.
A co-evolving proposer-critic RL framework improves GUI grounding accuracy by letting the model critique its own proposals rendered on screenshots.
Vision SR1 decomposes VLM reasoning into visual and language components and uses internal self-rewards to improve visual reasoning and reduce hallucinations more efficiently than external-supervision methods.
The survey organizes the shift of LLMs toward deliberate System 2 reasoning, covering model construction techniques, performance on math and coding benchmarks, and future research directions.
citing papers explorer
-
Variance-aware Reward Modeling with Anchor Guidance
Anchor-guided variance-aware reward modeling uses two response-level anchors to resolve non-identifiability in Gaussian models of pluralistic preferences, yielding provable identification, a joint training objective, and improved RLHF performance.
-
Latent Personality Alignment: Improving Harmlessness Without Mentioning Harms
LPA uses fewer than 100 personality trait statements to train LLMs for harmlessness, matching the robustness of methods using 150k+ harmful examples while generalizing better to new attacks.
-
Towards Robust LLM Post-Training: Automatic Failure Management for Reinforcement Fine-Tuning
Introduces the first benchmark for fine-grained failures in reinforcement fine-tuning of LLMs and an automatic management framework that detects, diagnoses, and remediates them.
-
General Preference Reinforcement Learning
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.
-
Diagnosing Training Inference Mismatch in LLM Reinforcement Learning
Training-inference mismatch in separated rollout and optimization stages of LLM RL can independently cause training collapse.
-
Reward Hacking in Rubric-Based Reinforcement Learning
Rubric-based RL verifiers can be gamed via partial criterion satisfaction and implicit-to-explicit tricks, yielding proxy gains that do not improve quality under rubric-free judges; stronger verifiers reduce but do not eliminate the mismatch.
-
Optimal Transport for LLM Reward Modeling from Noisy Preference
SelectiveRM applies optimal transport with a joint consistency discrepancy and partial mass relaxation to produce reward models that optimize a tighter upper bound on clean risk while autonomously dropping noisy preference samples.
-
Factored Causal Representation Learning for Robust Reward Modeling in RLHF
A factored causal representation learning method improves robustness of reward models in RLHF by isolating causal factors from biases like length and sycophancy using adversarial gradient reversal.
-
Supervising the search process produces reliable and generalizable information-seeking agents
Process supervision via RAG-Gym produces more reliable and generalizable search agents, with gains driven by higher-quality queries on out-of-domain multi-hop tasks.
-
Measure Twice, Click Once: Co-evolving Proposer and Visual Critic via Reinforcement Learning for GUI Grounding
A co-evolving proposer-critic RL framework improves GUI grounding accuracy by letting the model critique its own proposals rendered on screenshots.
-
Self-Rewarding Vision-Language Model via Reasoning Decomposition
Vision SR1 decomposes VLM reasoning into visual and language components and uses internal self-rewards to improve visual reasoning and reduce hallucinations more efficiently than external-supervision methods.
-
From System 1 to System 2: A Survey of Reasoning Large Language Models
The survey organizes the shift of LLMs toward deliberate System 2 reasoning, covering model construction techniques, performance on math and coding benchmarks, and future research directions.
- Rethinking the Comparison Unit in Sequence-Level Reinforcement Learning: An Equal-Length Paired Training Framework from Loss Correction to Sample Construction