A Tale of Tails: Model Collapse as a Change of Scaling Laws

Elvis Dohmatob, Yunzhen Feng, Pu Yang, Francois Charton, Julia Kempe
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:11165-11197, 2024.

Abstract

As AI model size grows, neural scaling laws have become a crucial tool to predict the improvements of large models when increasing capacity and the size of original (human or natural) training data. Yet, the widespread use of popular models means that the ecosystem of online data and text will co-evolve to progressively contain increased amounts of synthesized data. In this paper we ask: How will the scaling laws change in the inevitable regime where synthetic data makes its way into the training corpus? Will future models, still improve, or be doomed to degenerate up to total (model) collapse? We develop a theoretical framework of model collapse through the lens of scaling laws. We discover a wide range of decay phenomena, analyzing loss of scaling, shifted scaling with number of generations, the ”un-learning" of skills, and grokking when mixing human and synthesized data. Our theory is validated by large-scale experiments with a transformer on an arithmetic task and text generation using the large language model Llama2.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-dohmatob24b, title = {A Tale of Tails: Model Collapse as a Change of Scaling Laws}, author = {Dohmatob, Elvis and Feng, Yunzhen and Yang, Pu and Charton, Francois and Kempe, Julia}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {11165--11197}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/dohmatob24b/dohmatob24b.pdf}, url = {https://proceedings.mlr.press/v235/dohmatob24b.html}, abstract = {As AI model size grows, neural scaling laws have become a crucial tool to predict the improvements of large models when increasing capacity and the size of original (human or natural) training data. Yet, the widespread use of popular models means that the ecosystem of online data and text will co-evolve to progressively contain increased amounts of synthesized data. In this paper we ask: How will the scaling laws change in the inevitable regime where synthetic data makes its way into the training corpus? Will future models, still improve, or be doomed to degenerate up to total (model) collapse? We develop a theoretical framework of model collapse through the lens of scaling laws. We discover a wide range of decay phenomena, analyzing loss of scaling, shifted scaling with number of generations, the ”un-learning" of skills, and grokking when mixing human and synthesized data. Our theory is validated by large-scale experiments with a transformer on an arithmetic task and text generation using the large language model Llama2.} }
Endnote
%0 Conference Paper %T A Tale of Tails: Model Collapse as a Change of Scaling Laws %A Elvis Dohmatob %A Yunzhen Feng %A Pu Yang %A Francois Charton %A Julia Kempe %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-dohmatob24b %I PMLR %P 11165--11197 %U https://proceedings.mlr.press/v235/dohmatob24b.html %V 235 %X As AI model size grows, neural scaling laws have become a crucial tool to predict the improvements of large models when increasing capacity and the size of original (human or natural) training data. Yet, the widespread use of popular models means that the ecosystem of online data and text will co-evolve to progressively contain increased amounts of synthesized data. In this paper we ask: How will the scaling laws change in the inevitable regime where synthetic data makes its way into the training corpus? Will future models, still improve, or be doomed to degenerate up to total (model) collapse? We develop a theoretical framework of model collapse through the lens of scaling laws. We discover a wide range of decay phenomena, analyzing loss of scaling, shifted scaling with number of generations, the ”un-learning" of skills, and grokking when mixing human and synthesized data. Our theory is validated by large-scale experiments with a transformer on an arithmetic task and text generation using the large language model Llama2.
APA
Dohmatob, E., Feng, Y., Yang, P., Charton, F. & Kempe, J.. (2024). A Tale of Tails: Model Collapse as a Change of Scaling Laws. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:11165-11197 Available from https://proceedings.mlr.press/v235/dohmatob24b.html.

Related Material