Best-of-N through the Smoothing Lens: KL Divergence and Regret Analysis

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Aminian, Gholamali, Shenfeld, Idan, Asadi, Amir R., Beirami, Ahmad, Mroueh, Youssef
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908440942608384
author Aminian, Gholamali
Shenfeld, Idan
Asadi, Amir R.
Beirami, Ahmad
Mroueh, Youssef
author_facet Aminian, Gholamali
Shenfeld, Idan
Asadi, Amir R.
Beirami, Ahmad
Mroueh, Youssef
contents A simple yet effective method for inference-time alignment of generative models is Best-of-$N$ (BoN), where $N$ outcomes are sampled from a reference policy, evaluated using a proxy reward model, and the highest-scoring one is selected. While prior work argues that BoN is almost optimal in reward vs KL tradeoffs, the effectiveness of BoN depends critically on the quality of the proxy reward model used for selection. For this purpose, we study BoN through a smooth version known as Soft Best-of-N (SBoN) and develop a theoretical framework to address this gap. We analyze the scaling behaviour of BoN by providing bounds on the KL divergence between the SBoN policy and the reference policy, offering insights into how performance varies with the number of samples. We also study the regret gap, i.e., the gap between the expected true reward under the optimal policy and the SBoN policy. Our theoretical and empirical findings show that smoothing helps SBoN mitigate reward overoptimization, especially when the quality of the proxy reward is low.
format Preprint
id arxiv_https___arxiv_org_abs_2507_05913
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Best-of-N through the Smoothing Lens: KL Divergence and Regret Analysis
Aminian, Gholamali
Shenfeld, Idan
Asadi, Amir R.
Beirami, Ahmad
Mroueh, Youssef
Machine Learning
A simple yet effective method for inference-time alignment of generative models is Best-of-$N$ (BoN), where $N$ outcomes are sampled from a reference policy, evaluated using a proxy reward model, and the highest-scoring one is selected. While prior work argues that BoN is almost optimal in reward vs KL tradeoffs, the effectiveness of BoN depends critically on the quality of the proxy reward model used for selection. For this purpose, we study BoN through a smooth version known as Soft Best-of-N (SBoN) and develop a theoretical framework to address this gap. We analyze the scaling behaviour of BoN by providing bounds on the KL divergence between the SBoN policy and the reference policy, offering insights into how performance varies with the number of samples. We also study the regret gap, i.e., the gap between the expected true reward under the optimal policy and the SBoN policy. Our theoretical and empirical findings show that smoothing helps SBoN mitigate reward overoptimization, especially when the quality of the proxy reward is low.
title Best-of-N through the Smoothing Lens: KL Divergence and Regret Analysis
topic Machine Learning
url https://arxiv.org/abs/2507.05913