Reverse-KL Regularized Contextual Bandits for Robust Multimodal Alignment Against Reward Hacking

Abstract

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: Does the reverse-KL regularized contextual bandit formulation improve robustness against reward hacking in multimodal alignment tasks compared to existing offline preference learning methods. Direct Preference Optimization (DPO) has recently emerged as a popular approach to improve reinforcement learning with human feedback (RLHF), leading to better techniques to fine-tune large language models (LLM). A weakness of DPO, however, lies in its lack of capability to. 9 claims were extracted from source literature; 8 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 8.2/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research Question

Does the reverse-KL regularized contextual bandit formulation improve robustness against reward hacking in multimodal alignment tasks compared to existing offline preference learning methods?

Verification Level

Descriptive public verification status only; aggregate claim counts are public, but individual claim records are not exposed here.

Quality Tier

Descriptive public triage only; this tier does not alter current publication or DOI behavior.

Quality Dimensions

Automated triage signals derived from public fields; not human peer review or independent validation.

Correction Record

Public corrections are additive records. Current status does not claim the synthesis is error-free.

Provenance