The paper introduces RAG-Pref, a novel, training-free Retrieval Augmented Generation (RAG) method for preference alignment that significantly improves LLM refusal guardrails against agentic attacks wi…

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 SAVE, a framework that uses on-policy feedback and the value function to self-supervise and improve reward models, significantly enhancing RLHF performance across multiple benchma…

SafeSteer proposes a localized on-policy distillation method that restricts safety alignment to specific safety tokens, thereby achieving strong safety performance with minimal degradation to general…

The OISD framework improves language model reasoning by distilling on-policy predictive signals from the final output layer to intermediate representations, leading to substantial improvements on math…

The paper proposes CTRL-STEER, a closed-loop framework that adaptively adjusts intervention strength to stabilize concept regulation and improve task success in Vision-Language-Action models without r…

The paper introduces Expected Value Alignment (EVA), a novel reward modeling procedure that allows continuous scoring of intermediate reasoning steps in formal mathematics verification while maintaini…

The paper introduces AdvCL, a framework that repurposes adversarial perturbations as a geometric control signal to stabilize continual learning in large language models, significantly reducing forgett…

The paper proposes a novel, explicitly exploratory iterative Nash Learning from Human Feedback (NLHF) algorithm that achieves strong regret bounds for optimizing LLMs based on complex, non-scalar huma…

The paper introduces and evaluates five parameter alignment strategies that significantly mitigate catastrophic forgetting when continually pretraining multilingual expert language models across multi…

EchoRL proposes a lightweight module to exploit valuable learning signals from advantage-degenerated rollouts in Reinforcement Learning with Verifiable Rewards (RLVR), significantly improving LLM post…

The paper proposes DOA, a training-free attention policy that leverages self-attention in decoder-only SpeechLLMs to achieve high-quality, low-latency simultaneous long-form translation without requir…

DriftQL introduces a novel, efficient offline RL method that combines a drift-based behavioral regularizer with critic-driven policy improvement, achieving state-of-the-art performance while maintaini…

The paper introduces Prompted Policy Optimization (PromptPO), an LLM-based method that successfully optimizes policies for various sequential RL tasks, demonstrating that LLMs can replace classical RL…

The paper introduces the Markov decision contest, a new framework for reinforcement learning using pairwise preferences, and proves that stationary Markov policies are optimal and solvable efficiently…

The paper introduces Drifting Preference Optimization (DrPO), an efficient online method for preference finetuning one-step text-to-image generators that avoids complex gradient calculations and model…

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…

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…