Aligning Crowd Feedback via Distributional Preference Reward Modeling

TL;DR

This paper introduces DPRM, a novel framework that models diverse human preferences using a distributional approach and Bayesian updating, improving large language model alignment with broader societal expectations.

Contribution

The paper presents DPRM, a distributional preference reward model that captures diverse preferences and adapts to shifts, enhancing LLM alignment with population-wide expectations.

Findings

DPRM improves LLM response alignment with diverse human preferences.

DPRM yields more unbiased and contextually appropriate responses.

Experimental results show significant enhancement in model alignment quality.

Abstract

Deep Reinforcement Learning is widely used for aligning Large Language Models (LLM) with human preference. However, the conventional reward modelling is predominantly dependent on human annotations provided by a select cohort of individuals. Such dependence may unintentionally result in skewed models that reflect the inclinations of these annotators, thereby failing to adequately represent the wider population's expectations. We propose the Distributional Preference Reward Model (DPRM), a simple yet effective framework to align large language models with diverse human preferences. To this end, we characterize multiple preferences by a categorical distribution and introduce a Bayesian updater to accommodate shifted or new preferences. On top of that, we design an optimal-transportation-based loss to calibrate DPRM to align with the preference distribution. Finally, the expected reward is utilized to fine-tune an LLM policy to generate responses favoured by the population. Our experiments show that DPRM significantly enhances the alignment of LLMs with population preference, yielding more accurate, unbiased, and contextually appropriate responses.