SIGMA: Scalable Spectral Insights for LLM Model Collapse

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Gu, Yi, Pang, Lingyou, Ye, Xiangkun, Wang, Tianyu, Lin, Jianyu, Priebe, Carey E., Aue, Alexander
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917356021743616
author Gu, Yi
Pang, Lingyou
Ye, Xiangkun
Wang, Tianyu
Lin, Jianyu
Priebe, Carey E.
Aue, Alexander
author_facet Gu, Yi
Pang, Lingyou
Ye, Xiangkun
Wang, Tianyu
Lin, Jianyu
Priebe, Carey E.
Aue, Alexander
contents The rapid adoption of synthetic data for training Large Language Models (LLMs) has introduced the technical challenge of "model collapse"-a degenerative process where recursive training on model-generated content leads to a contraction of distributional variance and representational quality. While the phenomenology of collapse is increasingly evident, rigorous methods to quantify and predict its onset in high-dimensional spaces remain elusive. In this paper, we introduce SIGMA (Spectral Inequalities for Gram Matrix Analysis), a unified framework that benchmarks model collapse through the spectral lens of the embedding Gram matrix. By deriving and utilizing deterministic and stochastic bounds on the matrix's spectrum, SIGMA provides a mathematically grounded metric to track the contraction of the representation space. Crucially, our stochastic formulation enables scalable estimation of these bounds, making the framework applicable to large-scale foundation models where full eigendecomposition is intractable. We demonstrate that SIGMA effectively captures the transition towards degenerate states, offering both theoretical insights into the mechanics of collapse and a practical, scalable tool for monitoring the health of recursive training pipelines.
format Preprint
id arxiv_https___arxiv_org_abs_2601_03385
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle SIGMA: Scalable Spectral Insights for LLM Model Collapse
Gu, Yi
Pang, Lingyou
Ye, Xiangkun
Wang, Tianyu
Lin, Jianyu
Priebe, Carey E.
Aue, Alexander
Machine Learning
Probability
68T50(Primary), 60F05(Secondary)
I.2.7; G.3
The rapid adoption of synthetic data for training Large Language Models (LLMs) has introduced the technical challenge of "model collapse"-a degenerative process where recursive training on model-generated content leads to a contraction of distributional variance and representational quality. While the phenomenology of collapse is increasingly evident, rigorous methods to quantify and predict its onset in high-dimensional spaces remain elusive. In this paper, we introduce SIGMA (Spectral Inequalities for Gram Matrix Analysis), a unified framework that benchmarks model collapse through the spectral lens of the embedding Gram matrix. By deriving and utilizing deterministic and stochastic bounds on the matrix's spectrum, SIGMA provides a mathematically grounded metric to track the contraction of the representation space. Crucially, our stochastic formulation enables scalable estimation of these bounds, making the framework applicable to large-scale foundation models where full eigendecomposition is intractable. We demonstrate that SIGMA effectively captures the transition towards degenerate states, offering both theoretical insights into the mechanics of collapse and a practical, scalable tool for monitoring the health of recursive training pipelines.
title SIGMA: Scalable Spectral Insights for LLM Model Collapse
topic Machine Learning
Probability
68T50(Primary), 60F05(Secondary)
I.2.7; G.3
url https://arxiv.org/abs/2601.03385