Interpretable Reward Modeling with Active Concept Bottlenecks

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Laguna, Sonia, Kobalczyk, Katarzyna, Vogt, Julia E., Van der Schaar, Mihaela
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912492469354496
author Laguna, Sonia
Kobalczyk, Katarzyna
Vogt, Julia E.
Van der Schaar, Mihaela
author_facet Laguna, Sonia
Kobalczyk, Katarzyna
Vogt, Julia E.
Van der Schaar, Mihaela
contents We introduce Concept Bottleneck Reward Models (CB-RM), a reward modeling framework that enables interpretable preference learning through selective concept annotation. Unlike standard RLHF methods that rely on opaque reward functions, CB-RM decomposes reward prediction into human-interpretable concepts. To make this framework efficient in low-supervision settings, we formalize an active learning strategy that dynamically acquires the most informative concept labels. We propose an acquisition function based on Expected Information Gain and show that it significantly accelerates concept learning without compromising preference accuracy. Evaluated on the UltraFeedback dataset, our method outperforms baselines in interpretability and sample efficiency, marking a step towards more transparent, auditable, and human-aligned reward models.
format Preprint
id arxiv_https___arxiv_org_abs_2507_04695
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Interpretable Reward Modeling with Active Concept Bottlenecks
Laguna, Sonia
Kobalczyk, Katarzyna
Vogt, Julia E.
Van der Schaar, Mihaela
Machine Learning
We introduce Concept Bottleneck Reward Models (CB-RM), a reward modeling framework that enables interpretable preference learning through selective concept annotation. Unlike standard RLHF methods that rely on opaque reward functions, CB-RM decomposes reward prediction into human-interpretable concepts. To make this framework efficient in low-supervision settings, we formalize an active learning strategy that dynamically acquires the most informative concept labels. We propose an acquisition function based on Expected Information Gain and show that it significantly accelerates concept learning without compromising preference accuracy. Evaluated on the UltraFeedback dataset, our method outperforms baselines in interpretability and sample efficiency, marking a step towards more transparent, auditable, and human-aligned reward models.
title Interpretable Reward Modeling with Active Concept Bottlenecks
topic Machine Learning
url https://arxiv.org/abs/2507.04695