Structured Knowledge Accumulation: The Principle of Entropic Least Action in Forward-Only Neural Learning

Bouarfa Mahi Quantiota · 2025

Read Structured Knowledge Accumulation: The Principle of Entropic Least Action in Forward-Only Neural Learning by Bouarfa Mahi Quantiota (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-2504-03214.txt.

Description

Bouarfa Mahi Quantiota extends a learning framework that treats neural network training as a flow through time, in a 2025 paper. Most training runs in jumps: a step size times a count of rounds. The framework, called Structured Knowledge Accumulation, instead views learning as smooth motion, where the step size plays the role of a time step.

Two ideas anchor the extension. First, learning stays the same when the product of step size and rounds is fixed, which reveals a built-in timescale for each network. Second, a new measure, the Tensor Net function, ties choice odds, entropy slopes, and knowledge change into one gauge. Where that gauge crosses zero, the system rests, which gives a principled rule for when to stop training. The dynamics also obey a least-action law, like physical systems.

Catalog record

CREATORBouarfa Mahi Quantiota
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-03214

Browse the full library · About the Tower of Babel