Learning with Boolean threshold functions

Veit Elser, Manish Krishan Lal · 2026

Read Learning with Boolean threshold functions by Veit Elser, Manish Krishan Lal (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-2602-17493.txt.

Description

Veit Elser led a two-author team on learning with Boolean threshold functions. The team trains neural nets on Boolean data, where each node holds strictly plus or minus one. Trained models tend to match nets with nonzero weights also plus or minus one. Loss chasing gives way to nonconvex constraints. Each node runs a Boolean threshold function, and training splits in two through divide-and-concur.

One constraint keeps each node locally sound across inputs, weights, output. The other forces network-wide accord, lining neuron outputs up with later inputs and holding weights equal across net copies. The reflect-reflect-relax projection algorithm settles the pair. With a big enough margin floor, learned forms turn sparse and match nets of plain logical gates. Tests on cellular automata learning, binary autoencoding, logic network inference, and multiplier circuit discovery give exact answers or strong generalization where gradient methods falter.

Catalog record

CREATORVeit Elser, Manish Krishan Lal
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE96 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-17493

Browse the full library · About the Tower of Babel