Adiabatic replay for continual learning

Alexander Krawczyk, Alexander Gepperth

Primary Area: transfer learning, meta learning, and lifelong learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: continual learning ; replay ; mixture models

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: We propose a latent replay method for continual learning whose time complexity is independent of the number of tasks, and which outperforms existing deep generative replay approaches.

Abstract:

To avoid catastrophic forgetting, replay-based approaches to continual learning (CL) require, for each learning phase with new data, the replay of samples representing all of the previously learned knowledge. Since this knowledge grows over time, replay approaches invest linearly growing computational resources just re-learning what is already known. In this proof-of-concept study, we propose a generative replay-based CL strategy that we term adiabatic replay (AR), which achieves CL in constant time and memory complexity by making use of the (very common) situation where each new learning phase is adiabatic, i.e., represents only a small addition to existing knowledge.
AR owes its efficiency to the selective replay of samples that are similar to newly arriving ones. Indiscriminate replay is not required since AR is based on Gaussian Mixture Models (GMMs), which are capable of selectively updating their internal representation without catastrophic forgetting. Thus, the amount of to-be-replayed samples need not to depend on the amount of previously acquired knowledge at all. Based on the challenging CIFAR, SVHN and Fruits datasets in combination with foundation models, we confirm AR's superior scaling behavior while showing better accuracy than deep generative replay using VAEs and vanilla experience replay.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 3444