TY - GEN
T1 - Continual Learning via Dynamic Programming
AU - Krishnan, R.
AU - Balaprakash, Prasanna
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - 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.
AB - 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.
KW - Bellman principle
KW - continual learning
KW - dynamic programming
UR - http://www.scopus.com/inward/record.url?scp=85143637876&partnerID=8YFLogxK
U2 - 10.1109/ICPR56361.2022.9956042
DO - 10.1109/ICPR56361.2022.9956042
M3 - Conference contribution
AN - SCOPUS:85143637876
T3 - Proceedings - International Conference on Pattern Recognition
SP - 1350
EP - 1356
BT - 2022 26th International Conference on Pattern Recognition, ICPR 2022
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 26th International Conference on Pattern Recognition, ICPR 2022
Y2 - 21 August 2022 through 25 August 2022
ER -