Rational ANOVA Networks

Jusheng Zhang, Ningyuan Liu, Qinhan Lyu, Jing Yang, Keze Wang · 2026

Read Rational ANOVA Networks by Jusheng Zhang, Ningyuan Liu, Qinhan Lyu, Jing Yang, Keze Wang (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-2602-04006.txt.

Description

Networks built from interpretable pieces are Jusheng Zhang and 4 co-authors' answer to black-box deep learning. Their 2026 paper proposes the Rational-ANOVA Network, which decomposes a function into main effects and sparse pairwise interactions, following functional ANOVA. Each piece is a learnable rational function, a ratio of polynomials.

A strictly positive denominator prevents poles and blow-ups while still capturing sharp transitions. The low-order structure keeps the model readable and data-efficient. Rational pieces also extrapolate beyond the training range better than polynomial ones. Benchmarks on controlled functions and image classification test whether interpretability and accuracy can advance together.

Catalog record

CREATORJusheng Zhang, Ningyuan Liu, Qinhan Lyu, Jing Yang, Keze Wang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE92 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-04006

Browse the full library · About the Tower of Babel