Associative Memory and Generative Diffusion in the Zero-noise Limit

Joshua Hess [email protected], Quaid Morris [email protected] · 2025

Read Associative Memory and Generative Diffusion in the Zero-noise Limit by Joshua Hess [email protected], Quaid Morris [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-2506-05178.txt.

Introduction

Joshua Hess and Quaid Morris prove that generative diffusion models and associative memories are two views of one machine, meeting where the noise dies. Diffusion models build data by slowly removing noise. Associative memories settle into stored patterns. Their 2025 paper shows the first converges to the second as noise vanishes, and it characterizes stability, robustness, and capacity for both.

A class of smooth dynamical systems turns out to approximate any associative memory. Diffusion is then that system plus a whisper of white noise. As noise fades, the system's shifts between creating and remembering happen at isolated, countable tipping points. The math gives both fields a shared language. Learning theorists get a bridge with load-bearing proofs, not metaphors.

Catalog record

CREATORJoshua Hess [email protected], Quaid Morris [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE178 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-05178

Browse the full library · About the Tower of Babel