Abstract
Continual learning (CL) algorithms seek to train a model when faced with similar tasks observed in a sequential manner. Despite promising methodological advancements, however, there is a lack of theoretical frameworks that enable analysis of learning challenges such as generalization and catastrophic forgetting, especially in applications where the tasks are generated continuously through a partial differential equation. To address this lack, we present a new theoretical framework that models the learning dynamics in CL through dynamic programming. Using the proposed framework, we derive a new method that adopts stochastic-gradient-driven alternating optimization to balance generalization and catastrophic forgetting. We establish conditions for convergence and show that, on CL benchmark datasets, our method achieves accuracies better than or comparable to those of existing state-of-the-art methods.
Original language | English |
---|---|
Title of host publication | 2022 26th International Conference on Pattern Recognition, ICPR 2022 |
Publisher | Institute of Electrical and Electronics Engineers Inc. |
Pages | 1350-1356 |
Number of pages | 7 |
ISBN (Electronic) | 9781665490627 |
DOIs | |
State | Published - 2022 |
Externally published | Yes |
Event | 26th International Conference on Pattern Recognition, ICPR 2022 - Montreal, Canada Duration: Aug 21 2022 → Aug 25 2022 |
Publication series
Name | Proceedings - International Conference on Pattern Recognition |
---|---|
Volume | 2022-August |
ISSN (Print) | 1051-4651 |
Conference
Conference | 26th International Conference on Pattern Recognition, ICPR 2022 |
---|---|
Country/Territory | Canada |
City | Montreal |
Period | 08/21/22 → 08/25/22 |
Funding
ACKNOWLEDGMENT This work was supported by the U.S. Department of Energy, Office of Science, Advanced Scientific Computing Research, under Contract DE-AC02-06CH11357 and a DOE Early Career Research Program award. We are grateful for the computing resources from the Joint Laboratory for System Evaluation and Leadership Computing Facility at Argonne.
Keywords
- Bellman principle
- continual learning
- dynamic programming