FAMOSE: A ReAct Approach to Automated Feature Discovery

Keith Burghardt, Jienan Liu, Sadman Sakib, Yuning Hao, Bo Li · 2026

Read FAMOSE: A ReAct Approach to Automated Feature Discovery by Keith Burghardt, Jienan Liu, Sadman Sakib, Yuning Hao, Bo Li (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-2602-17641.txt.

Description

Keith Burghardt and team introduce FAMOSE, an agent system for automated feature discovery in machine learning. Feature engineering stays a key bottleneck, above all for tabular data, where the space of candidate features is vast and domain expertise is costly. FAMOSE is billed as the first agentic ReAct system aimed at this task.

The agent explores, builds and refines features on its own, while selection and evaluation tools sit inside its loop. Tests put it at or near the state of the art on classification, and at the state of the art on regression, with small average gains on both. It also proves more robust to errors than rival algorithms. The authors credit the ReAct loop: its context log of what worked and what failed steers the model toward fresher, more inventive features.

Catalog record

CREATORKeith Burghardt, Jienan Liu, Sadman Sakib, Yuning Hao, Bo Li
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-17641

Browse the full library · About the Tower of Babel