Scalable Pre-training of Large Autoregressive Image Models

Alaaeldin El-Nouby, Michal Klein, Shuangfei Zhai, Miguel Ángel Bautista, Vaishaal Shankar, Alexander T Toshev, Joshua M. Susskind, Armand Joulin
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:12371-12384, 2024.

Abstract

This paper introduces AIM, a collection of vision models pre-trained with an autoregressive objective. These models are inspired by their textual counterparts, i.e., Large Language Models (LLMs), and exhibit similar scaling properties. Specifically, we highlight two key findings: (1) the performance of the visual features scale with both the model capacity and the quantity of data, (2) the value of the objective function correlates with the performance of the model on downstream tasks. We illustrate the practical implication of these findings by pre-training a 7 billion parameter AIM on 2 billion images, that achieves 84.0% on ImageNet-1k with a frozen trunk. Interestingly, even at this scale, we observe no sign of saturation in performance, suggesting that AIM potentially represents a new frontier for training large-scale vision models. The pre-training of AIM is similar to the pre-training of LLMs, and does not require any image-specific strategy to stabilize the training at scale.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-el-nouby24a, title = {Scalable Pre-training of Large Autoregressive Image Models}, author = {El-Nouby, Alaaeldin and Klein, Michal and Zhai, Shuangfei and Bautista, Miguel \'{A}ngel and Shankar, Vaishaal and Toshev, Alexander T and Susskind, Joshua M. and Joulin, Armand}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {12371--12384}, 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/el-nouby24a/el-nouby24a.pdf}, url = {https://proceedings.mlr.press/v235/el-nouby24a.html}, abstract = {This paper introduces AIM, a collection of vision models pre-trained with an autoregressive objective. These models are inspired by their textual counterparts, i.e., Large Language Models (LLMs), and exhibit similar scaling properties. Specifically, we highlight two key findings: (1) the performance of the visual features scale with both the model capacity and the quantity of data, (2) the value of the objective function correlates with the performance of the model on downstream tasks. We illustrate the practical implication of these findings by pre-training a 7 billion parameter AIM on 2 billion images, that achieves 84.0% on ImageNet-1k with a frozen trunk. Interestingly, even at this scale, we observe no sign of saturation in performance, suggesting that AIM potentially represents a new frontier for training large-scale vision models. The pre-training of AIM is similar to the pre-training of LLMs, and does not require any image-specific strategy to stabilize the training at scale.} }
Endnote
%0 Conference Paper %T Scalable Pre-training of Large Autoregressive Image Models %A Alaaeldin El-Nouby %A Michal Klein %A Shuangfei Zhai %A Miguel Ángel Bautista %A Vaishaal Shankar %A Alexander T Toshev %A Joshua M. Susskind %A Armand Joulin %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-el-nouby24a %I PMLR %P 12371--12384 %U https://proceedings.mlr.press/v235/el-nouby24a.html %V 235 %X This paper introduces AIM, a collection of vision models pre-trained with an autoregressive objective. These models are inspired by their textual counterparts, i.e., Large Language Models (LLMs), and exhibit similar scaling properties. Specifically, we highlight two key findings: (1) the performance of the visual features scale with both the model capacity and the quantity of data, (2) the value of the objective function correlates with the performance of the model on downstream tasks. We illustrate the practical implication of these findings by pre-training a 7 billion parameter AIM on 2 billion images, that achieves 84.0% on ImageNet-1k with a frozen trunk. Interestingly, even at this scale, we observe no sign of saturation in performance, suggesting that AIM potentially represents a new frontier for training large-scale vision models. The pre-training of AIM is similar to the pre-training of LLMs, and does not require any image-specific strategy to stabilize the training at scale.
APA
El-Nouby, A., Klein, M., Zhai, S., Bautista, M.Á., Shankar, V., Toshev, A.T., Susskind, J.M. & Joulin, A.. (2024). Scalable Pre-training of Large Autoregressive Image Models. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:12371-12384 Available from https://proceedings.mlr.press/v235/el-nouby24a.html.

Related Material