Cross-Modal Fine-Tuning: Align then Refine

Junhong Shen, Liam Li, Lucio M. Dery, Corey Staten, Mikhail Khodak, Graham Neubig, Ameet Talwalkar
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:31030-31056, 2023.

Abstract

Fine-tuning large-scale pretrained models has led to tremendous progress in well-studied modalities such as vision and NLP. However, similar gains have not been observed in many other modalities due to a lack of relevant pretrained models. In this work, we propose ORCA, a general cross-modal fine-tuning framework that extends the applicability of a single large-scale pretrained model to diverse modalities. ORCA adapts to a target task via an align-then-refine workflow: given the target input, ORCA first learns an embedding network that aligns the embedded feature distribution with the pretraining modality. The pretrained model is then fine-tuned on the embedded data to exploit the knowledge shared across modalities. Through extensive experiments, we show that ORCA obtains state-of-the-art results on 3 benchmarks containing over 60 datasets from 12 modalities, outperforming a wide range of hand-designed, AutoML, general-purpose, and task-specific cross-modal methods. We highlight the importance of data alignment via a series of ablation studies and exemplify ORCA’s utility in data-limited regimes.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-shen23e, title = {Cross-Modal Fine-Tuning: Align then Refine}, author = {Shen, Junhong and Li, Liam and Dery, Lucio M. and Staten, Corey and Khodak, Mikhail and Neubig, Graham and Talwalkar, Ameet}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {31030--31056}, 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/shen23e/shen23e.pdf}, url = {https://proceedings.mlr.press/v202/shen23e.html}, abstract = {Fine-tuning large-scale pretrained models has led to tremendous progress in well-studied modalities such as vision and NLP. However, similar gains have not been observed in many other modalities due to a lack of relevant pretrained models. In this work, we propose ORCA, a general cross-modal fine-tuning framework that extends the applicability of a single large-scale pretrained model to diverse modalities. ORCA adapts to a target task via an align-then-refine workflow: given the target input, ORCA first learns an embedding network that aligns the embedded feature distribution with the pretraining modality. The pretrained model is then fine-tuned on the embedded data to exploit the knowledge shared across modalities. Through extensive experiments, we show that ORCA obtains state-of-the-art results on 3 benchmarks containing over 60 datasets from 12 modalities, outperforming a wide range of hand-designed, AutoML, general-purpose, and task-specific cross-modal methods. We highlight the importance of data alignment via a series of ablation studies and exemplify ORCA’s utility in data-limited regimes.} }
Endnote
%0 Conference Paper %T Cross-Modal Fine-Tuning: Align then Refine %A Junhong Shen %A Liam Li %A Lucio M. Dery %A Corey Staten %A Mikhail Khodak %A Graham Neubig %A Ameet Talwalkar %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-shen23e %I PMLR %P 31030--31056 %U https://proceedings.mlr.press/v202/shen23e.html %V 202 %X Fine-tuning large-scale pretrained models has led to tremendous progress in well-studied modalities such as vision and NLP. However, similar gains have not been observed in many other modalities due to a lack of relevant pretrained models. In this work, we propose ORCA, a general cross-modal fine-tuning framework that extends the applicability of a single large-scale pretrained model to diverse modalities. ORCA adapts to a target task via an align-then-refine workflow: given the target input, ORCA first learns an embedding network that aligns the embedded feature distribution with the pretraining modality. The pretrained model is then fine-tuned on the embedded data to exploit the knowledge shared across modalities. Through extensive experiments, we show that ORCA obtains state-of-the-art results on 3 benchmarks containing over 60 datasets from 12 modalities, outperforming a wide range of hand-designed, AutoML, general-purpose, and task-specific cross-modal methods. We highlight the importance of data alignment via a series of ablation studies and exemplify ORCA’s utility in data-limited regimes.
APA
Shen, J., Li, L., Dery, L.M., Staten, C., Khodak, M., Neubig, G. & Talwalkar, A.. (2023). Cross-Modal Fine-Tuning: Align then Refine. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:31030-31056 Available from https://proceedings.mlr.press/v202/shen23e.html.

Related Material