Data-Efficient Contrastive Language-Image Pretraining: Prioritizing Data Quality over Quantity

Siddharth Joshi, Arnav Jain, Ali Payani, Baharan Mirzasoleiman
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:1000-1008, 2024.

Abstract

Contrastive Language-Image Pre-training (CLIP) on large-scale image-caption datasets learns representations that can achieve remarkable zero-shot generalization. However, such models require a massive amount of pre-training data. Improving the quality of the pre-training data has been shown to be much more effective in improving CLIP’s performance than increasing its volume. Nevertheless, finding small subsets of training data that provably generalize best has remained an open question. In this work, we propose the first theoretically rigorous data selection method for CLIP. We show that subsets that closely preserve the cross-covariance of the images and captions of the full data provably achieve a superior generalization performance.Our extensive experiments on ConceptualCaptions3M and ConceptualCaptions12M demonstrate that subsets found by \textsc{ClipCov} achieve over 2.7x and 1.4x the accuracy of the next best baseline on ImageNet and its shifted versions. Moreover, we show that our subsets obtain 1.5x the average accuracy across 11 downstream datasets, of the next best baseline. The code is available at: \url{https://github.com/BigML-CS-UCLA/clipcov-data-efficient-clip}.

Cite this Paper


BibTeX
@InProceedings{pmlr-v238-joshi24a, title = {Data-Efficient Contrastive Language-Image Pretraining: Prioritizing Data Quality over Quantity}, author = {Joshi, Siddharth and Jain, Arnav and Payani, Ali and Mirzasoleiman, Baharan}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {1000--1008}, year = {2024}, editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen}, volume = {238}, series = {Proceedings of Machine Learning Research}, month = {02--04 May}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v238/joshi24a/joshi24a.pdf}, url = {https://proceedings.mlr.press/v238/joshi24a.html}, abstract = {Contrastive Language-Image Pre-training (CLIP) on large-scale image-caption datasets learns representations that can achieve remarkable zero-shot generalization. However, such models require a massive amount of pre-training data. Improving the quality of the pre-training data has been shown to be much more effective in improving CLIP’s performance than increasing its volume. Nevertheless, finding small subsets of training data that provably generalize best has remained an open question. In this work, we propose the first theoretically rigorous data selection method for CLIP. We show that subsets that closely preserve the cross-covariance of the images and captions of the full data provably achieve a superior generalization performance.Our extensive experiments on ConceptualCaptions3M and ConceptualCaptions12M demonstrate that subsets found by \textsc{ClipCov} achieve over 2.7x and 1.4x the accuracy of the next best baseline on ImageNet and its shifted versions. Moreover, we show that our subsets obtain 1.5x the average accuracy across 11 downstream datasets, of the next best baseline. The code is available at: \url{https://github.com/BigML-CS-UCLA/clipcov-data-efficient-clip}.} }
Endnote
%0 Conference Paper %T Data-Efficient Contrastive Language-Image Pretraining: Prioritizing Data Quality over Quantity %A Siddharth Joshi %A Arnav Jain %A Ali Payani %A Baharan Mirzasoleiman %B Proceedings of The 27th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2024 %E Sanjoy Dasgupta %E Stephan Mandt %E Yingzhen Li %F pmlr-v238-joshi24a %I PMLR %P 1000--1008 %U https://proceedings.mlr.press/v238/joshi24a.html %V 238 %X Contrastive Language-Image Pre-training (CLIP) on large-scale image-caption datasets learns representations that can achieve remarkable zero-shot generalization. However, such models require a massive amount of pre-training data. Improving the quality of the pre-training data has been shown to be much more effective in improving CLIP’s performance than increasing its volume. Nevertheless, finding small subsets of training data that provably generalize best has remained an open question. In this work, we propose the first theoretically rigorous data selection method for CLIP. We show that subsets that closely preserve the cross-covariance of the images and captions of the full data provably achieve a superior generalization performance.Our extensive experiments on ConceptualCaptions3M and ConceptualCaptions12M demonstrate that subsets found by \textsc{ClipCov} achieve over 2.7x and 1.4x the accuracy of the next best baseline on ImageNet and its shifted versions. Moreover, we show that our subsets obtain 1.5x the average accuracy across 11 downstream datasets, of the next best baseline. The code is available at: \url{https://github.com/BigML-CS-UCLA/clipcov-data-efficient-clip}.
APA
Joshi, S., Jain, A., Payani, A. & Mirzasoleiman, B.. (2024). Data-Efficient Contrastive Language-Image Pretraining: Prioritizing Data Quality over Quantity. Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:1000-1008 Available from https://proceedings.mlr.press/v238/joshi24a.html.

Related Material