Improving Set Function Approximation with Quasi-Arithmetic Neural Networks

Tomas Tokar, Scott Sanner · 2026

Read Improving Set Function Approximation with Quasi-Arithmetic Neural Networks by Tomas Tokar, Scott Sanner (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-2602-04941.txt.

Description

Tomas Tokar and Scott Sanner wrote this 2026 machine learning paper, due to appear at ICLR 2026, about learning from sets. Sets have no natural order, so models like DeepSets squash them with a fixed operation such as a sum or a maximum. A fixed squash can blur useful information and limits what the model can express.

The authors propose a learnable replacement: a neural version of the Kolmogorov mean, which averages values after passing them through an invertible function. Networks built around it, called QUANNs, are shown to approximate a broad class of set functions in theory. In tests they beat leading baselines across varied benchmarks, and their learned representations transferred well even to tasks with no sets at all.

Catalog record

CREATORTomas Tokar, Scott Sanner
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE77 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-04941

Browse the full library · About the Tower of Babel