The paper establishes the first tilde O(epsilon^{-1}) upper bounds and matching lower bounds for forward-KL-regularized offline contextual bandits under single-policy concentrability in both tabular and general function approximation settings.
hub
Advances in Neural Information Processing Systems , volume=
25 Pith papers cite this work. Polarity classification is still indexing.
25
Pith papers citing it
hub tools
representative citing papers
The cumulative token IS ratio gives unbiased prefix correction and lower variance than full-sequence ratios for token-level gradients in LLM policy optimization, enabling CTPO to outperform GRPO and GSPO baselines on mathematical reasoning tasks.
o1-like models overthink easy tasks; self-training reduces compute use without accuracy loss on GSM8K, MATH500, GPQA, and AIME.
Meta Agent Search uses a meta-agent to iteratively program novel agentic systems in code, producing agents that outperform state-of-the-art hand-designed ones across coding, science, and math while transferring across domains and models.
COALA applies convex optimization reformulations of neural networks to direct preference optimization, claiming single-GPU training with ~18% of DPO's TFLOPs and competitive performance on multiple datasets and models up to 8B parameters.
dFlowGRPO is a new rate-aware RL method for discrete flow models that outperforms prior GRPO approaches on image generation and matches continuous flow models while supporting broad probability paths.
CAGE uses common-agency games and an EPEC algorithm to compute equilibrium policies that balance multiple conflicting objectives for test-time LLM alignment.
Pre-training provides a geometric warm start in a single-index model that enables weak-to-strong generalization up to a supervisor-limited bound, with empirical phase-transition evidence in LLMs.
ZAYA1-8B is a reasoning MoE model with 700M active parameters that matches larger models on math and coding benchmarks and reaches 91.9% on AIME'25 via Markovian RSA test-time compute.
REGLU guides LoRA-based unlearning via representation subspaces and orthogonal regularization to outperform prior methods on forget-retain trade-off in LLM benchmarks.
A score-ranking loss enables controllable summarization by aligning outputs to evaluation scores, matching SOTA performance with dimension-specific control on LLaMA, Qwen, and Mistral.
Muon optimizer with weight decay and update scaling achieves ~2x efficiency over AdamW for large LLMs, shown via the Moonlight 3B/16B MoE model trained on 5.7T tokens.
LIMO achieves 63.3% on AIME24 and 95.6% on MATH500 via supervised fine-tuning on roughly 1% of the data used by prior models, supporting the claim that minimal strategic examples suffice when pre-training has already encoded domain knowledge.
PRIME enables online process reward model updates in LLM RL using implicit rewards from rollouts and outcome labels, yielding 15.1% average gains on reasoning benchmarks and surpassing a stronger instruct model with 10% of the data.
SCoRe uses multi-turn online RL with regularization on self-generated traces to improve LLM self-correction, achieving 15.6% and 9.1% gains on MATH and HumanEval for Gemini models.
RLAIF matches RLHF on summarization and dialogue tasks, with a direct-RLAIF variant achieving superior results by using LLM rewards directly during training.
LLMs assigned high or low status personas in multi-turn dialogues exhibit socio-cognitive effects including language coordination, pronoun patterns, persuasion success, and compliance with unsafe requests.
ARGUS uses a Prosecutor-Defender-Umpire multi-agent setup plus RAG and chain-of-thought rewards to adapt ad policy enforcement to new regulations using minimal fresh labels.
A reference-free proxy scoring framework combined with GIRB calibration produces better-aligned evaluation metrics for summarization and outperforms baselines across seven datasets.
APMPO boosts average Pass@1 scores on math reasoning benchmarks by 3 points over GRPO by using an adaptive power-mean policy objective and feedback-driven clipping bounds in RLVR training.
FREIA applies free energy principles and adaptive advantage shaping to unsupervised RL, outperforming baselines by 0.5-3.5 Pass@1 points on math reasoning with a 1.5B model.
VANGUARD is a staged-training VLM framework that reports 94% ROC-AUC and 84% F1 on UCF-Crime while adding chain-of-thought reasoning and spatial grounding to video anomaly detection.
SmolLM2 is a 1.7B-parameter language model that outperforms Qwen2.5-1.5B and Llama3.2-1B after overtraining on 11 trillion tokens using custom FineMath, Stack-Edu, and SmolTalk datasets in a multi-stage pipeline.
DeepSight uses parallel latent feature prediction in BEV for long-horizon world modeling and adaptive text reasoning to reach state-of-the-art closed-loop performance on the Bench2drive benchmark.
citing papers explorer
-
Fast Rates for Offline Contextual Bandits with Forward-KL Regularization under Single-Policy Concentrability
The paper establishes the first tilde O(epsilon^{-1}) upper bounds and matching lower bounds for forward-KL-regularized offline contextual bandits under single-policy concentrability in both tabular and general function approximation settings.
-
Rethinking Importance Sampling in LLM Policy Optimization: A Cumulative Token Perspective
The cumulative token IS ratio gives unbiased prefix correction and lower variance than full-sequence ratios for token-level gradients in LLM policy optimization, enabling CTPO to outperform GRPO and GSPO baselines on mathematical reasoning tasks.
-
Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs
o1-like models overthink easy tasks; self-training reduces compute use without accuracy loss on GSM8K, MATH500, GPQA, and AIME.
-
Automated Design of Agentic Systems
Meta Agent Search uses a meta-agent to iteratively program novel agentic systems in code, producing agents that outperform state-of-the-art hand-designed ones across coding, science, and math while transferring across domains and models.
-
Convex Optimization for Alignment and Preference Learning on a Single GPU
COALA applies convex optimization reformulations of neural networks to direct preference optimization, claiming single-GPU training with ~18% of DPO's TFLOPs and competitive performance on multiple datasets and models up to 8B parameters.
-
dFlowGRPO: Rate-Aware Policy Optimization for Discrete Flow Models
dFlowGRPO is a new rate-aware RL method for discrete flow models that outperforms prior GRPO approaches on image generation and matches continuous flow models while supporting broad probability paths.
-
Common-agency Games for Multi-Objective Test-Time Alignment
CAGE uses common-agency games and an EPEC algorithm to compute equilibrium policies that balance multiple conflicting objectives for test-time LLM alignment.
-
On the Blessing of Pre-training in Weak-to-Strong Generalization
Pre-training provides a geometric warm start in a single-index model that enables weak-to-strong generalization up to a supervisor-limited bound, with empirical phase-transition evidence in LLMs.
-
ZAYA1-8B Technical Report
ZAYA1-8B is a reasoning MoE model with 700M active parameters that matches larger models on math and coding benchmarks and reaches 91.9% on AIME'25 via Markovian RSA test-time compute.
-
Representation-Guided Parameter-Efficient LLM Unlearning
REGLU guides LoRA-based unlearning via representation subspaces and orthogonal regularization to outperform prior methods on forget-retain trade-off in LLM benchmarks.
-
Learning to Control Summaries with Score Ranking
A score-ranking loss enables controllable summarization by aligning outputs to evaluation scores, matching SOTA performance with dimension-specific control on LLaMA, Qwen, and Mistral.
-
Muon is Scalable for LLM Training
Muon optimizer with weight decay and update scaling achieves ~2x efficiency over AdamW for large LLMs, shown via the Moonlight 3B/16B MoE model trained on 5.7T tokens.
-
LIMO: Less is More for Reasoning
LIMO achieves 63.3% on AIME24 and 95.6% on MATH500 via supervised fine-tuning on roughly 1% of the data used by prior models, supporting the claim that minimal strategic examples suffice when pre-training has already encoded domain knowledge.
-
Process Reinforcement through Implicit Rewards
PRIME enables online process reward model updates in LLM RL using implicit rewards from rollouts and outcome labels, yielding 15.1% average gains on reasoning benchmarks and surpassing a stronger instruct model with 10% of the data.
-
Training Language Models to Self-Correct via Reinforcement Learning
SCoRe uses multi-turn online RL with regularization on self-generated traces to improve LLM self-correction, achieving 15.6% and 9.1% gains on MATH and HumanEval for Gemini models.
-
RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedback
RLAIF matches RLHF on summarization and dialogue tasks, with a direct-RLAIF variant achieving superior results by using LLM rewards directly during training.
-
Do LLM Agents Mirror Socio-Cognitive Effects in Power-Asymmetric Conversations?
LLMs assigned high or low status personas in multi-turn dialogues exhibit socio-cognitive effects including language coordination, pronoun patterns, persuasion success, and compliance with unsafe requests.
-
ARGUS: Policy-Adaptive Ad Governance via Evolving Reinforcement with Adversarial Umpiring
ARGUS uses a Prosecutor-Defender-Umpire multi-agent setup plus RAG and chain-of-thought rewards to adapt ad policy enforcement to new regulations using minimal fresh labels.
-
Calibrating Model-Based Evaluation Metrics for Summarization
A reference-free proxy scoring framework combined with GIRB calibration produces better-aligned evaluation metrics for summarization and outperforms baselines across seven datasets.
-
Adapt to Thrive! Adaptive Power-Mean Policy Optimization for Improved LLM Reasoning
APMPO boosts average Pass@1 scores on math reasoning benchmarks by 3 points over GRPO by using an adaptive power-mean policy objective and feedback-driven clipping bounds in RLVR training.
-
Free Energy-Driven Reinforcement Learning with Adaptive Advantage Shaping for Unsupervised Reasoning in LLMs
FREIA applies free energy principles and adaptive advantage shaping to unsupervised RL, outperforming baselines by 0.5-3.5 Pass@1 points on math reasoning with a 1.5B model.
-
Reasoning-Guided Grounding: Elevating Video Anomaly Detection through Multimodal Large Language Models
VANGUARD is a staged-training VLM framework that reports 94% ROC-AUC and 84% F1 on UCF-Crime while adding chain-of-thought reasoning and spatial grounding to video anomaly detection.
-
SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model
SmolLM2 is a 1.7B-parameter language model that outperforms Qwen2.5-1.5B and Llama3.2-1B after overtraining on 11 trillion tokens using custom FineMath, Stack-Edu, and SmolTalk datasets in a multi-stage pipeline.
-
DeepSight: Long-Horizon World Modeling via Latent States Prediction for End-to-End Autonomous Driving
DeepSight uses parallel latent feature prediction in BEV for long-horizon world modeling and adaptive text reasoning to reach state-of-the-art closed-loop performance on the Bench2drive benchmark.
-
A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence
The paper delivers the first systematic review of self-evolving agents, structured around what components evolve, when adaptation occurs, and how it is implemented.