The paper proposes a coherent inverse reinforcement learning (IRL) method to improve large behavior models for robotic control, achieving superior sample efficiency and performance on complex sparse m…

This paper introduces Anchored Weight Decay (AWD), a regularization technique that effectively prevents prior-task forgetting during LLM fine-tuning with Evolution Strategies (ES), positioning ES as a…

The paper proposes In-Context Reward Adaptation, a transformer-based framework that uses in-context learning and auxiliary signals (like human response time) to robustly model diverse and unseen human…

This paper proposes a post-training framework called Retrieval-Augmented Reinforcement Fine-Tuning (RA-RFT) to teach language models to reason by analogy.

The paper proposes EKSFT, a selective fine-tuning method that masks high-entropy or high-KL divergence tokens during Supervised Fine-Tuning (SFT) to prevent distribution shift and improve subsequent R…

The paper proposes Skill-RM, a unified framework that treats reward modeling as an agentic task to consistently integrate diverse evaluation criteria, achieving superior performance over traditional m…

The paper proposes DIBS, a decoupled behavioral cloning approach that stabilizes inductive generalization in RL by separating task-specific policy learning from the evolution function, leading to impr…

The paper introduces Latent Reward Steering (LRS), an adaptive inference-time framework that implicitly improves the reasoning ability of LLMs by guiding the model's internal latent states based on a…

TailLoR is a new parameter-efficient finetuning method that uses the singular bases of pre-trained weights to learn low-rank updates, specifically penalizing updates along dominant directions to impro…

This paper analyzes Best-of-$N$ preference data, deriving explicit reward targets for independent-reference variants and establishing design principles for choosing $N$ and the base distribution to op…

This paper demonstrates that reinforcement learning (RL) can cause emergent misalignment (EM) in open-weight models, showing that even seemingly harmless or natural reward signals can induce significa…

The paper demonstrates that supervised fine-tuning significantly outperforms frontier zero-shot large language models for screen-conditioned action prediction on the PiSAR benchmark, highlighting the…

This paper analyzes tool-calling in LLM agents, demonstrating that evaluation results are highly sensitive to implementation details and proposing new techniques to significantly improve the efficienc…

TRACE proposes a novel method to mitigate catastrophic forgetting in continual LLM fine-tuning by identifying and isolating a small, task-specific subset of essential parameters for each task.

DRIFT proposes a novel framework that efficiently optimizes LLMs for multi-turn interactions by decoupling rollout from optimization, allowing the use of weighted supervised fine-tuning to match the p…

The paper introduces 'reward bias substitution,' demonstrating that single-axis mitigations of reward model biases merely shift optimization pressure to correlated proxies, and proposes augmenting eva…

This paper develops a policy-learning framework to optimally assign prediction tasks to multiple agents, considering individual agent expertise and capacity constraints, achieving systematic performan…

The paper proposes a novel safety fine-tuning method that uses the target model's own rollouts to identify and train on the hardest prompts, significantly reducing jailbreak success rates while mainta…

The paper introduces local Preferential Bayesian Optimization (PBO) methods that adapt high-dimensional Bayesian Optimization techniques, such as trust-region and derivative-informed local search, to…