$\pi$-Tuning: Transferring Multimodal Foundation Models with Optimal Multi-task Interpolation

Chengyue Wu, Teng Wang, Yixiao Ge, Zeyu Lu, Ruisong Zhou, Ying Shan, Ping Luo
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:37713-37727, 2023.

Abstract

Foundation models have achieved great advances in multi-task learning with a unified interface of unimodal and multimodal tasks. However, the potential of such multi-task learners has not been exploited during transfer learning. In this work, we present a universal parameter-efficient transfer learning method, termed Predict-Interpolate Tuning ($\pi$-Tuning), for vision, language, and vision-language tasks. It aggregates the parameters of lightweight task-specific experts learned from similar tasks to aid the target downstream task. The task similarities are predicted in a unified modality-independent space, yielding a scalable graph to demonstrate task relationships. $\pi$-Tuning has several appealing benefits. First, it flexibly explores both intra- and inter-modal transferability between similar tasks to improve the accuracy and robustness of transfer learning, especially in data-scarce scenarios. Second, it offers a systematical solution for transfer learning with multi-task prediction-and-then-interpolation, compatible with diverse types of parameter-efficient experts, such as prompt and adapter. Third, an extensive study of task-level mutual benefits on 14 unimodal and 6 multimodal datasets shows that $\pi$-Tuning surpasses fine-tuning and other parameter-efficient transfer learning methods both in full-shot and low-shot regimes. The task graph also enables an in-depth interpretable analysis of task transferability across modalities. The code will be available at https://github.com/TencentARC/pi-Tuning.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-wu23t, title = {$\pi$-Tuning: Transferring Multimodal Foundation Models with Optimal Multi-task Interpolation}, author = {Wu, Chengyue and Wang, Teng and Ge, Yixiao and Lu, Zeyu and Zhou, Ruisong and Shan, Ying and Luo, Ping}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {37713--37727}, 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/wu23t/wu23t.pdf}, url = {https://proceedings.mlr.press/v202/wu23t.html}, abstract = {Foundation models have achieved great advances in multi-task learning with a unified interface of unimodal and multimodal tasks. However, the potential of such multi-task learners has not been exploited during transfer learning. In this work, we present a universal parameter-efficient transfer learning method, termed Predict-Interpolate Tuning ($\pi$-Tuning), for vision, language, and vision-language tasks. It aggregates the parameters of lightweight task-specific experts learned from similar tasks to aid the target downstream task. The task similarities are predicted in a unified modality-independent space, yielding a scalable graph to demonstrate task relationships. $\pi$-Tuning has several appealing benefits. First, it flexibly explores both intra- and inter-modal transferability between similar tasks to improve the accuracy and robustness of transfer learning, especially in data-scarce scenarios. Second, it offers a systematical solution for transfer learning with multi-task prediction-and-then-interpolation, compatible with diverse types of parameter-efficient experts, such as prompt and adapter. Third, an extensive study of task-level mutual benefits on 14 unimodal and 6 multimodal datasets shows that $\pi$-Tuning surpasses fine-tuning and other parameter-efficient transfer learning methods both in full-shot and low-shot regimes. The task graph also enables an in-depth interpretable analysis of task transferability across modalities. The code will be available at https://github.com/TencentARC/pi-Tuning.} }
Endnote
%0 Conference Paper %T $\pi$-Tuning: Transferring Multimodal Foundation Models with Optimal Multi-task Interpolation %A Chengyue Wu %A Teng Wang %A Yixiao Ge %A Zeyu Lu %A Ruisong Zhou %A Ying Shan %A Ping Luo %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-wu23t %I PMLR %P 37713--37727 %U https://proceedings.mlr.press/v202/wu23t.html %V 202 %X Foundation models have achieved great advances in multi-task learning with a unified interface of unimodal and multimodal tasks. However, the potential of such multi-task learners has not been exploited during transfer learning. In this work, we present a universal parameter-efficient transfer learning method, termed Predict-Interpolate Tuning ($\pi$-Tuning), for vision, language, and vision-language tasks. It aggregates the parameters of lightweight task-specific experts learned from similar tasks to aid the target downstream task. The task similarities are predicted in a unified modality-independent space, yielding a scalable graph to demonstrate task relationships. $\pi$-Tuning has several appealing benefits. First, it flexibly explores both intra- and inter-modal transferability between similar tasks to improve the accuracy and robustness of transfer learning, especially in data-scarce scenarios. Second, it offers a systematical solution for transfer learning with multi-task prediction-and-then-interpolation, compatible with diverse types of parameter-efficient experts, such as prompt and adapter. Third, an extensive study of task-level mutual benefits on 14 unimodal and 6 multimodal datasets shows that $\pi$-Tuning surpasses fine-tuning and other parameter-efficient transfer learning methods both in full-shot and low-shot regimes. The task graph also enables an in-depth interpretable analysis of task transferability across modalities. The code will be available at https://github.com/TencentARC/pi-Tuning.
APA
Wu, C., Wang, T., Ge, Y., Lu, Z., Zhou, R., Shan, Y. & Luo, P.. (2023). $\pi$-Tuning: Transferring Multimodal Foundation Models with Optimal Multi-task Interpolation. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:37713-37727 Available from https://proceedings.mlr.press/v202/wu23t.html.

Related Material