Decision Trees That Remember: Gradient-Based Learning of Recurrent Decision Trees with Memory

Sascha MartonUniversity of Mannheim [email protected], Moritz SchneiderBoehringer Ingelheim [email protected] · 2025

Read Decision Trees That Remember: Gradient-Based Learning of Recurrent Decision Trees with Memory by Sascha MartonUniversity of Mannheim [email protected], Moritz SchneiderBoehringer Ingelheim [email protected] (2025) 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-2502-04052.txt.

Description

Sascha Marton and Moritz Schneider give decision trees a memory for sequence data in a 2025 paper. Trees shine on tidy tables but miss patterns that unfold over time. Neural networks learn such patterns well, yet their rules stay hidden and hard to audit.

Their ReMeDe trees carry an inner state, updated by crisp split rules like a classic tree. Gradient methods tune those hard rules directly, a trick that keeps the model readable. Output and state updates both flow from the same split logic. Early tests on made-up benchmarks show the design can learn long-range links in sequences. Wider trials lie ahead.

Catalog record

CREATORSascha MartonUniversity of Mannheim [email protected], Moritz SchneiderBoehringer Ingelheim [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE32 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-04052

Browse the full library · About the Tower of Babel