University of Illinois Urbana-Champaign

A Theoretical Basis for Model Collapse in Recursive Training

Borkar, Vivek S.

Permalink

https://hdl.handle.net/2142/130253

Description

Title
A Theoretical Basis for Model Collapse in Recursive Training
Author(s)
Borkar, Vivek S.
Issue Date
2025-09-17
Keyword(s)
  • Generative models
  • Recursive training
  • Model collapse
  • Convergence of probability measures
  • Martingale convergence theorem
Abstract
It is known that recursive training from generative models can lead to the so called ‘collapse’ of the simulated probability distribution. This note shows that one in fact gets two different asymptotic behaviours depending on whether an external source, howsoever minor, is also contributing samples.
Publisher
Allerton Conference on Communication, Control, and Computing
Series/Report Name or Number
2025 61st Allerton Conference on Communication, Control, and Computing Proceedings
ISSN
2836-4503
Type of Resource
Text
Genre of Resource
Conference Paper/Presentation
Language
eng
Handle URL
https://hdl.handle.net/2142/130253&&
Copyright and License Information
Copyright 2025 is held by Vivek S. Borkar.

Owning Collections

61st Allerton Conference - 2025 PRIMARY
Allerton Conference on Communication, Control, and Computing, 1977-present