Task Confusion and Catastrophic Forgetting in Class-Incremental Learning: A Mathematical Framework for Discriminative and Generative Modelings
Abstract
In class-incremental learning (class-IL), models must classify all previously seen classes at test time without task-IDs, leading to task confusion. Despite being a key challenge, task confusion lacks a theoretical understanding. We present a novel mathematical framework for class-IL and prove the Infeasibility Theorem, showing optimal class-IL is impossible with discriminative modeling due to task confusion. However, we establish the Feasibility Theorem, demonstrating that generative modeling can achieve optimal class-IL by overcoming task confusion. We then assess popular class-IL strategies, including regularization, bias-correction, replay, and generative classifier, using our framework. Our analysis suggests that adopting generative modeling, either for generative replay or direct classification (generative classifier), is essential for optimal class-IL.
Cite
Text
Nori and Kim. "Task Confusion and Catastrophic Forgetting in Class-Incremental Learning: A Mathematical Framework for Discriminative and Generative Modelings." Neural Information Processing Systems, 2024. doi:10.52202/079017-1510Markdown
[Nori and Kim. "Task Confusion and Catastrophic Forgetting in Class-Incremental Learning: A Mathematical Framework for Discriminative and Generative Modelings." Neural Information Processing Systems, 2024.](https://mlanthology.org/neurips/2024/nori2024neurips-task/) doi:10.52202/079017-1510BibTeX
@inproceedings{nori2024neurips-task,
title = {{Task Confusion and Catastrophic Forgetting in Class-Incremental Learning: A Mathematical Framework for Discriminative and Generative Modelings}},
author = {Nori, Milad Khademi and Kim, II-Min},
booktitle = {Neural Information Processing Systems},
year = {2024},
doi = {10.52202/079017-1510},
url = {https://mlanthology.org/neurips/2024/nori2024neurips-task/}
}