One Goal, Many Challenges: Robust Preference Optimization Amid Content-Aware and Multi-Source Noise

Amirabbas Afzali, Amirhossein Afsharrad, Seyed Shahabeddin Mousavi, Sanjay Lall

Keywords: Reinforcement Learning from Human Feedback, Preference Alignment, Backdoor Attack, Robust Preference Optimization

TL;DR: This paper presents CNRPO, a novel framework that mitigates content-dependent biases in preference learning, improving alignment with primary human preferences while controlling for secondary biases like response length and harmfulness.

Abstract: Large Language Models (LLMs) have made significant strides in generating human-like responses, largely due to preference alignment techniques. However, these methods often assume unbiased human feedback, which is rarely the case in real-world scenarios. This paper introduces Content-Aware Noise-Resilient Preference Optimization (CNRPO), a novel framework that addresses multiple sources of content-dependent noise in preference learning. CNRPO employs a multi-objective optimization approach to separate true preferences from content-aware noises, effectively mitigating their impact. We leverage backdoor attack mechanisms to efficiently learn and control various noise sources within a single model. Theoretical analysis and extensive experiments on different synthetic noisy datasets demonstrate that CNRPO significantly improves alignment with primary human preferences while controlling for secondary noises and biases, such as response length and harmfulness.

Submission Number: 298