Continual Learning Beyond a Single Model

Thang Doan, Seyed Iman Mirzadeh, Mehrdad Farajtabar
Proceedings of The 2nd Conference on Lifelong Learning Agents, PMLR 232:961-991, 2023.

Abstract

A growing body of research in continual learning focuses on the catastrophic forgetting problem. While many attempts have been made to alleviate this problem, the majority of the methods assume a \textit{single model} in the continual learning setup. In this work, we question this assumption and show that employing \textit{ensemble models} can be a simple yet effective method to improve continual performance. However, ensembles’ training and inference costs can increase significantly as the number of models grows. Motivated by this limitation, we study different ensemble models to understand their benefits and drawbacks in continual learning scenarios. Finally, to overcome the high compute cost of ensembles, we leverage recent advances in neural network subspace to propose a computationally cheap algorithm with similar runtime to a single model yet enjoying the performance benefits of ensembles.

Cite this Paper

BibTeX
@InProceedings{pmlr-v232-doan23a, title = {Continual Learning Beyond a Single Model}, author = {Doan, Thang and Mirzadeh, Seyed Iman and Farajtabar, Mehrdad}, booktitle = {Proceedings of The 2nd Conference on Lifelong Learning Agents}, pages = {961--991}, year = {2023}, editor = {Chandar, Sarath and Pascanu, Razvan and Sedghi, Hanie and Precup, Doina}, volume = {232}, series = {Proceedings of Machine Learning Research}, month = {22--25 Aug}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v232/doan23a/doan23a.pdf}, url = {https://proceedings.mlr.press/v232/doan23a.html}, abstract = {A growing body of research in continual learning focuses on the catastrophic forgetting problem. While many attempts have been made to alleviate this problem, the majority of the methods assume a \textit{single model} in the continual learning setup. In this work, we question this assumption and show that employing \textit{ensemble models} can be a simple yet effective method to improve continual performance. However, ensembles’ training and inference costs can increase significantly as the number of models grows. Motivated by this limitation, we study different ensemble models to understand their benefits and drawbacks in continual learning scenarios. Finally, to overcome the high compute cost of ensembles, we leverage recent advances in neural network subspace to propose a computationally cheap algorithm with similar runtime to a single model yet enjoying the performance benefits of ensembles.} }
Endnote
%0 Conference Paper %T Continual Learning Beyond a Single Model %A Thang Doan %A Seyed Iman Mirzadeh %A Mehrdad Farajtabar %B Proceedings of The 2nd Conference on Lifelong Learning Agents %C Proceedings of Machine Learning Research %D 2023 %E Sarath Chandar %E Razvan Pascanu %E Hanie Sedghi %E Doina Precup %F pmlr-v232-doan23a %I PMLR %P 961--991 %U https://proceedings.mlr.press/v232/doan23a.html %V 232 %X A growing body of research in continual learning focuses on the catastrophic forgetting problem. While many attempts have been made to alleviate this problem, the majority of the methods assume a \textit{single model} in the continual learning setup. In this work, we question this assumption and show that employing \textit{ensemble models} can be a simple yet effective method to improve continual performance. However, ensembles’ training and inference costs can increase significantly as the number of models grows. Motivated by this limitation, we study different ensemble models to understand their benefits and drawbacks in continual learning scenarios. Finally, to overcome the high compute cost of ensembles, we leverage recent advances in neural network subspace to propose a computationally cheap algorithm with similar runtime to a single model yet enjoying the performance benefits of ensembles.
APA
Doan, T., Mirzadeh, S.I. & Farajtabar, M.. (2023). Continual Learning Beyond a Single Model. Proceedings of The 2nd Conference on Lifelong Learning Agents, in Proceedings of Machine Learning Research 232:961-991 Available from https://proceedings.mlr.press/v232/doan23a.html.

Related Material