Analyzing Diffusion as Serial Reproduction

Raja Marjieh, Ilia Sucholutsky, Thomas A Langlois, Nori Jacoby, Thomas L. Griffiths
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:24005-24019, 2023.

Abstract

Diffusion models are a class of generative models that learn to synthesize samples by inverting a diffusion process that gradually maps data into noise. While these models have enjoyed great success recently, a full theoretical understanding of their observed properties is still lacking, in particular, their weak sensitivity to the choice of noise family and the role of adequate scheduling of noise levels for good synthesis. By identifying a correspondence between diffusion models and a well-known paradigm in cognitive science known as serial reproduction, whereby human agents iteratively observe and reproduce stimuli from memory, we show how the aforementioned properties of diffusion models can be explained as a natural consequence of this correspondence. We then complement our theoretical analysis with simulations that exhibit these key features. Our work highlights how classic paradigms in cognitive science can shed light on state-of-the-art machine learning problems.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-marjieh23a, title = {Analyzing Diffusion as Serial Reproduction}, author = {Marjieh, Raja and Sucholutsky, Ilia and Langlois, Thomas A and Jacoby, Nori and Griffiths, Thomas L.}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {24005--24019}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/marjieh23a/marjieh23a.pdf}, url = {https://proceedings.mlr.press/v202/marjieh23a.html}, abstract = {Diffusion models are a class of generative models that learn to synthesize samples by inverting a diffusion process that gradually maps data into noise. While these models have enjoyed great success recently, a full theoretical understanding of their observed properties is still lacking, in particular, their weak sensitivity to the choice of noise family and the role of adequate scheduling of noise levels for good synthesis. By identifying a correspondence between diffusion models and a well-known paradigm in cognitive science known as serial reproduction, whereby human agents iteratively observe and reproduce stimuli from memory, we show how the aforementioned properties of diffusion models can be explained as a natural consequence of this correspondence. We then complement our theoretical analysis with simulations that exhibit these key features. Our work highlights how classic paradigms in cognitive science can shed light on state-of-the-art machine learning problems.} }
Endnote
%0 Conference Paper %T Analyzing Diffusion as Serial Reproduction %A Raja Marjieh %A Ilia Sucholutsky %A Thomas A Langlois %A Nori Jacoby %A Thomas L. Griffiths %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-marjieh23a %I PMLR %P 24005--24019 %U https://proceedings.mlr.press/v202/marjieh23a.html %V 202 %X Diffusion models are a class of generative models that learn to synthesize samples by inverting a diffusion process that gradually maps data into noise. While these models have enjoyed great success recently, a full theoretical understanding of their observed properties is still lacking, in particular, their weak sensitivity to the choice of noise family and the role of adequate scheduling of noise levels for good synthesis. By identifying a correspondence between diffusion models and a well-known paradigm in cognitive science known as serial reproduction, whereby human agents iteratively observe and reproduce stimuli from memory, we show how the aforementioned properties of diffusion models can be explained as a natural consequence of this correspondence. We then complement our theoretical analysis with simulations that exhibit these key features. Our work highlights how classic paradigms in cognitive science can shed light on state-of-the-art machine learning problems.
APA
Marjieh, R., Sucholutsky, I., Langlois, T.A., Jacoby, N. & Griffiths, T.L.. (2023). Analyzing Diffusion as Serial Reproduction. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:24005-24019 Available from https://proceedings.mlr.press/v202/marjieh23a.html.

Related Material