Overcoming Catastrophic Forgetting by Bayesian Generative Regularization
Pei-Hung Chen, Wei Wei, Cho-Jui Hsieh, Bo Dai
Abstract
In this paper, we propose a new method to over-come catastrophic forgetting by adding generative regularization to Bayesian inference frame-work. Bayesian method provides a general frame-work for continual learning. We could further construct a generative regularization term for all given classification models by leveraging energy-based models and Langevin dynamic sampling to enrich the features learned in each task. By combining discriminative and generative loss together, we empirically show that the proposed method outperforms state-of-the-art methods on a variety of tasks, avoiding catastrophic forgetting in continual learning. In particular, the proposed method outperforms baseline methods over 15%on the Fashion-MNIST dataset and 10%on the CUB dataset.
BibTeX
@InProceedings{pmlr-v139-chen21v,
title = {Overcoming Catastrophic Forgetting by Bayesian Generative Regularization},
author = {Chen, Pei-Hung and Wei, Wei and Hsieh, Cho-Jui and Dai, Bo},
booktitle = {Proceedings of the 38th International Conference on Machine Learning},
pages = {1760--1770},
year = {2021},
editor = {Meila, Marina and Zhang, Tong},
volume = {139},
series = {Proceedings of Machine Learning Research},
month = {18--24 Jul},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v139/chen21v/chen21v.pdf},
url = {https://proceedings.mlr.press/v139/chen21v.html},
abstract = {In this paper, we propose a new method to over-come catastrophic forgetting by adding generative regularization to Bayesian inference frame-work. Bayesian method provides a general frame-work for continual learning. We could further construct a generative regularization term for all given classification models by leveraging energy-based models and Langevin dynamic sampling to enrich the features learned in each task. By combining discriminative and generative loss together, we empirically show that the proposed method outperforms state-of-the-art methods on a variety of tasks, avoiding catastrophic forgetting in continual learning. In particular, the proposed method outperforms baseline methods over 15%on the Fashion-MNIST dataset and 10%on the CUB dataset.}
}