POUF: Prompt-Oriented Unsupervised Fine-tuning for Large Pre-trained Models

Korawat Tanwisuth, Shujian Zhang, Huangjie Zheng, Pengcheng He, Mingyuan Zhou
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:33816-33832, 2023.

Abstract

Through prompting, large-scale pre-trained models have become more expressive and powerful, gaining significant attention in recent years. Though these big models have zero-shot capabilities, in general, labeled data are still required to adapt them to downstream tasks. To overcome this critical limitation, we propose an unsupervised fine-tuning framework to directly fine-tune the model or prompt on the unlabeled target data. We demonstrate how to apply our method to both language-augmented vision and masked-language models, by aligning the discrete distributions extracted from the prompts and target data. To verify our approach’s applicability, we conduct extensive experiments on image classification, sentiment analysis, and natural language inference tasks. Across 13 image-related tasks and 15 language-related ones, the proposed approach achieves consistent improvements over the baselines. PyTorch code is available at https://github.com/korawat-tanwisuth/POUF.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-tanwisuth23a, title = {{POUF}: Prompt-Oriented Unsupervised Fine-tuning for Large Pre-trained Models}, author = {Tanwisuth, Korawat and Zhang, Shujian and Zheng, Huangjie and He, Pengcheng and Zhou, Mingyuan}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {33816--33832}, 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/tanwisuth23a/tanwisuth23a.pdf}, url = {https://proceedings.mlr.press/v202/tanwisuth23a.html}, abstract = {Through prompting, large-scale pre-trained models have become more expressive and powerful, gaining significant attention in recent years. Though these big models have zero-shot capabilities, in general, labeled data are still required to adapt them to downstream tasks. To overcome this critical limitation, we propose an unsupervised fine-tuning framework to directly fine-tune the model or prompt on the unlabeled target data. We demonstrate how to apply our method to both language-augmented vision and masked-language models, by aligning the discrete distributions extracted from the prompts and target data. To verify our approach’s applicability, we conduct extensive experiments on image classification, sentiment analysis, and natural language inference tasks. Across 13 image-related tasks and 15 language-related ones, the proposed approach achieves consistent improvements over the baselines. PyTorch code is available at https://github.com/korawat-tanwisuth/POUF.} }
Endnote
%0 Conference Paper %T POUF: Prompt-Oriented Unsupervised Fine-tuning for Large Pre-trained Models %A Korawat Tanwisuth %A Shujian Zhang %A Huangjie Zheng %A Pengcheng He %A Mingyuan Zhou %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-tanwisuth23a %I PMLR %P 33816--33832 %U https://proceedings.mlr.press/v202/tanwisuth23a.html %V 202 %X Through prompting, large-scale pre-trained models have become more expressive and powerful, gaining significant attention in recent years. Though these big models have zero-shot capabilities, in general, labeled data are still required to adapt them to downstream tasks. To overcome this critical limitation, we propose an unsupervised fine-tuning framework to directly fine-tune the model or prompt on the unlabeled target data. We demonstrate how to apply our method to both language-augmented vision and masked-language models, by aligning the discrete distributions extracted from the prompts and target data. To verify our approach’s applicability, we conduct extensive experiments on image classification, sentiment analysis, and natural language inference tasks. Across 13 image-related tasks and 15 language-related ones, the proposed approach achieves consistent improvements over the baselines. PyTorch code is available at https://github.com/korawat-tanwisuth/POUF.
APA
Tanwisuth, K., Zhang, S., Zheng, H., He, P. & Zhou, M.. (2023). POUF: Prompt-Oriented Unsupervised Fine-tuning for Large Pre-trained Models. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:33816-33832 Available from https://proceedings.mlr.press/v202/tanwisuth23a.html.

Related Material