Continual Learners are Incremental Model Generalizers

Jaehong Yoon, Sung Ju Hwang, Yue Cao
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:40129-40146, 2023.

Abstract

Motivated by the efficiency and rapid convergence of pre-trained models for solving downstream tasks, this paper extensively studies the impact of Continual Learning (CL) models as pre-trainers. We find that, in both supervised and unsupervised CL, the transfer quality of representations does not show a noticeable degradation of fine-tuning performance but rather increases gradually. This is because CL models can learn improved task-general features when easily forgetting task-specific knowledge. Based on this observation, we suggest a new unsupervised CL framework with masked modeling, which aims to capture fluent task-generic representation during training. Furthermore, we propose a new fine-tuning scheme, GLobal Attention Discretization (GLAD), that preserves rich task-generic representation during solving downstream tasks. The model fine-tuned with GLAD achieves competitive performance and can also be used as a good pre-trained model itself. We believe this paper breaks the barriers between pre-training and fine-tuning steps and leads to a sustainable learning framework in which the continual learner incrementally improves model generalization, yielding better transfer to unseen tasks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-yoon23b, title = {Continual Learners are Incremental Model Generalizers}, author = {Yoon, Jaehong and Hwang, Sung Ju and Cao, Yue}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {40129--40146}, 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/yoon23b/yoon23b.pdf}, url = {https://proceedings.mlr.press/v202/yoon23b.html}, abstract = {Motivated by the efficiency and rapid convergence of pre-trained models for solving downstream tasks, this paper extensively studies the impact of Continual Learning (CL) models as pre-trainers. We find that, in both supervised and unsupervised CL, the transfer quality of representations does not show a noticeable degradation of fine-tuning performance but rather increases gradually. This is because CL models can learn improved task-general features when easily forgetting task-specific knowledge. Based on this observation, we suggest a new unsupervised CL framework with masked modeling, which aims to capture fluent task-generic representation during training. Furthermore, we propose a new fine-tuning scheme, GLobal Attention Discretization (GLAD), that preserves rich task-generic representation during solving downstream tasks. The model fine-tuned with GLAD achieves competitive performance and can also be used as a good pre-trained model itself. We believe this paper breaks the barriers between pre-training and fine-tuning steps and leads to a sustainable learning framework in which the continual learner incrementally improves model generalization, yielding better transfer to unseen tasks.} }
Endnote
%0 Conference Paper %T Continual Learners are Incremental Model Generalizers %A Jaehong Yoon %A Sung Ju Hwang %A Yue Cao %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-yoon23b %I PMLR %P 40129--40146 %U https://proceedings.mlr.press/v202/yoon23b.html %V 202 %X Motivated by the efficiency and rapid convergence of pre-trained models for solving downstream tasks, this paper extensively studies the impact of Continual Learning (CL) models as pre-trainers. We find that, in both supervised and unsupervised CL, the transfer quality of representations does not show a noticeable degradation of fine-tuning performance but rather increases gradually. This is because CL models can learn improved task-general features when easily forgetting task-specific knowledge. Based on this observation, we suggest a new unsupervised CL framework with masked modeling, which aims to capture fluent task-generic representation during training. Furthermore, we propose a new fine-tuning scheme, GLobal Attention Discretization (GLAD), that preserves rich task-generic representation during solving downstream tasks. The model fine-tuned with GLAD achieves competitive performance and can also be used as a good pre-trained model itself. We believe this paper breaks the barriers between pre-training and fine-tuning steps and leads to a sustainable learning framework in which the continual learner incrementally improves model generalization, yielding better transfer to unseen tasks.
APA
Yoon, J., Hwang, S.J. & Cao, Y.. (2023). Continual Learners are Incremental Model Generalizers. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:40129-40146 Available from https://proceedings.mlr.press/v202/yoon23b.html.

Related Material