Deep Actor-Critics with Tight Risk Certificates

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Tasdighi, Bahareh, Haussmann, Manuel, Wu, Yi-Shan, Masegosa, Andres R., Kandemir, Melih
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911287916625920
author Tasdighi, Bahareh
Haussmann, Manuel
Wu, Yi-Shan
Masegosa, Andres R.
Kandemir, Melih
author_facet Tasdighi, Bahareh
Haussmann, Manuel
Wu, Yi-Shan
Masegosa, Andres R.
Kandemir, Melih
contents Deep actor-critic algorithms have reached a level where they influence everyday life. They are a driving force behind continual improvement of large language models through user feedback. However, their deployment in physical systems is not yet widely adopted, mainly because no validation scheme fully quantifies their risk of malfunction. We demonstrate that it is possible to develop tight risk certificates for deep actor-critic algorithms that predict generalization performance from validation-time observations. Our key insight centers on the effectiveness of minimal evaluation data. A small feasible set of evaluation roll-outs collected from a pretrained policy suffices to produce accurate risk certificates when combined with a simple adaptation of PAC-Bayes theory. Specifically, we adopt a recently introduced recursive PAC-Bayes approach, which splits validation data into portions and recursively builds PAC-Bayes bounds on the excess loss of each portion's predictor, using the predictor from the previous portion as a data-informed prior. Our empirical results across multiple locomotion tasks, actor-critic methods, and policy expertise levels demonstrate risk certificates tight enough to be considered for practical use.
format Preprint
id arxiv_https___arxiv_org_abs_2505_19682
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Deep Actor-Critics with Tight Risk Certificates
Tasdighi, Bahareh
Haussmann, Manuel
Wu, Yi-Shan
Masegosa, Andres R.
Kandemir, Melih
Machine Learning
Deep actor-critic algorithms have reached a level where they influence everyday life. They are a driving force behind continual improvement of large language models through user feedback. However, their deployment in physical systems is not yet widely adopted, mainly because no validation scheme fully quantifies their risk of malfunction. We demonstrate that it is possible to develop tight risk certificates for deep actor-critic algorithms that predict generalization performance from validation-time observations. Our key insight centers on the effectiveness of minimal evaluation data. A small feasible set of evaluation roll-outs collected from a pretrained policy suffices to produce accurate risk certificates when combined with a simple adaptation of PAC-Bayes theory. Specifically, we adopt a recently introduced recursive PAC-Bayes approach, which splits validation data into portions and recursively builds PAC-Bayes bounds on the excess loss of each portion's predictor, using the predictor from the previous portion as a data-informed prior. Our empirical results across multiple locomotion tasks, actor-critic methods, and policy expertise levels demonstrate risk certificates tight enough to be considered for practical use.
title Deep Actor-Critics with Tight Risk Certificates
topic Machine Learning
url https://arxiv.org/abs/2505.19682