Salvato in:
Dettagli Bibliografici
Autori principali: Dumoulin, Vincent, Johnson, Daniel D., Castro, Pablo Samuel, Larochelle, Hugo, Dauphin, Yann
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:https://arxiv.org/abs/2311.14115
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911753323937792
author Dumoulin, Vincent
Johnson, Daniel D.
Castro, Pablo Samuel
Larochelle, Hugo
Dauphin, Yann
author_facet Dumoulin, Vincent
Johnson, Daniel D.
Castro, Pablo Samuel
Larochelle, Hugo
Dauphin, Yann
contents Learning from human feedback (LHF) -- and in particular learning from pairwise preferences -- has recently become a crucial ingredient in training large language models (LLMs), and has been the subject of much research. Most recent works frame it as a reinforcement learning problem, where a reward function is learned from pairwise preference data and the LLM is treated as a policy which is adapted to maximize the rewards, often under additional regularization constraints. We propose an alternative interpretation which centers on the generative process for pairwise preferences and treats LHF as a density estimation problem. We provide theoretical and empirical results showing that for a family of generative processes defined via preference behavior distribution equations, training a reward function on pairwise preferences effectively models an annotator's implicit preference distribution. Finally, we discuss and present findings on "annotator misspecification" -- failure cases where wrong modeling assumptions are made about annotator behavior, resulting in poorly-adapted models -- suggesting that approaches that learn from pairwise human preferences could have trouble learning from a population of annotators with diverse viewpoints.
format Preprint
id arxiv_https___arxiv_org_abs_2311_14115
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle A density estimation perspective on learning from pairwise human preferences
Dumoulin, Vincent
Johnson, Daniel D.
Castro, Pablo Samuel
Larochelle, Hugo
Dauphin, Yann
Machine Learning
Artificial Intelligence
Computation and Language
Learning from human feedback (LHF) -- and in particular learning from pairwise preferences -- has recently become a crucial ingredient in training large language models (LLMs), and has been the subject of much research. Most recent works frame it as a reinforcement learning problem, where a reward function is learned from pairwise preference data and the LLM is treated as a policy which is adapted to maximize the rewards, often under additional regularization constraints. We propose an alternative interpretation which centers on the generative process for pairwise preferences and treats LHF as a density estimation problem. We provide theoretical and empirical results showing that for a family of generative processes defined via preference behavior distribution equations, training a reward function on pairwise preferences effectively models an annotator's implicit preference distribution. Finally, we discuss and present findings on "annotator misspecification" -- failure cases where wrong modeling assumptions are made about annotator behavior, resulting in poorly-adapted models -- suggesting that approaches that learn from pairwise human preferences could have trouble learning from a population of annotators with diverse viewpoints.
title A density estimation perspective on learning from pairwise human preferences
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2311.14115