English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

This paper studies the theoretical framework of the alignment process ofgenerative models with Reinforcement Learning from Human Feedback (RLHF). Weconsider a standard mathematical formulation, the reverse-KL regularizedcontextual bandit for RLHF. Despite its widespread practical application, arigorous theoretical analysis of this formulation remains open. We investigateits theoretical properties both in offline and online settings and proposeefficient algorithms with finite-sample theoretical guarantees. Our workbridges the gap between theory and practice by linking our theoretical insightswith existing practical alignment algorithms such as Direct PreferenceOptimization (DPO) and Rejection Sampling Optimization (RSO). Furthermore,these findings and connections also offer both theoretical and practicalcommunities new tools and insights for future algorithmic design of alignmentalgorithms.

Gibbs Sampling from Human Feedback: A Provable KL- constrained Framework for RLHF