A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Unknown
A survey of recent advances in using large language models as judges for evaluation, critique, preference modeling, and feedback on text and code.
Unknown
The paper identifies a hidden bias in Process Reward Models caused by severe imbalance in step-level feedback and proposes PRISM to reward the right reasoning.
Frank L. Lewis, Draguna Vrabie, Kyriakos G. Vamvoudakis
This paper unifies adaptive and optimal control using reinforcement learning to design optimal adaptive controllers for discrete- and continuous-time systems without full dynamics knowledge.
Xuanliang Zhang, Dingzirui Wang, Keyan Xu, et al.
Introduces Effective Feedback Compute (EFC), a trace-level scaling coordinate for agent harnesses that predicts performance better than raw compute and improves pass rates while reducing cost.
Ebenezer Gelo, Geraud Nangue Tasse, Steven James, et al.
Proposes RCI framework to convert sparse trajectory-level stop-feedback into dense per-step costs for safe offline RL, preserving feasible policy set and optimal Lagrangian.
Unknown
This paper proposes Softmax-DPO, a novel adaptation of Direct Preference Optimization for recommendation systems, integrating softmax loss to enhance implicit feedback modeling.
Unknown
This paper provides a comprehensive overview of RLHF, covering its core methods, challenges, and applications in building large-scale machine learning systems.
Unknown
Self-Play Preference Optimization (SPO) is a minimalist reinforcement learning from human feedback algorithm that avoids training a separate reward model by using self-play.
Unknown
This paper introduces variational preference learning (VPL), a method that personalizes RLHF by learning a latent user-specific preference distribution from diverse human feedback.
Yuntao Bai, Andy Jones, Kamal Ndousse, et al.
This paper applies preference modeling and reinforcement learning from human feedback to fine-tune language models to be helpful and harmless assistants.
Unknown
RLAIF scales reinforcement learning from human feedback by using AI feedback to train reward models, reducing reliance on human annotations.
Josef Dai, Xuehai Pan, Ruiyang Sun, et al.
Safe RLHF decouples reward modeling and safety constraints to train language models that are both helpful and safe.