Improving Regret Approximation for Unsupervised Dynamic Environment Generation

Harry Mead, Bruno Lacerda, Jakob Foerster, Nick Hawes · 2026

Read Improving Regret Approximation for Unsupervised Dynamic Environment Generation by Harry Mead, Bruno Lacerda, Jakob Foerster, Nick Hawes (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-14957.txt.

Description

Harry Mead, Bruno Lacerda, Jakob Foerster, and Nick Hawes work on training agents in worlds that design themselves. Their 2026 arXiv paper sits in unsupervised environment design. The goal is curricula that make reinforcement learning agents generalize and cope with new tasks.

Building a good curriculum is still hard. A few settings in a world's design space can sharply raise the skill a policy needs. Methods in this field lean on a regret estimate to pick the next task. The authors work on making that regret estimate closer to the truth.

Catalog record

CREATORHarry Mead, Bruno Lacerda, Jakob Foerster, Nick Hawes
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE79 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-14957

Browse the full library · About the Tower of Babel