Distillation Techniques for Pseudo-rehearsal Based Incremental Learning
This work addresses the problem of catastrophic forgetting for lifelong learning systems, but it is incremental as it builds on existing GAN-based methods with modest improvements.
The paper tackles catastrophic forgetting in incremental learning by addressing biases in GAN-based distillation methods, proposing a new approach using AC-GAN distillation that achieves performance comparable to state-of-the-art rehearsal-based methods on MNIST and CIFAR10 datasets.
The ability to learn from incrementally arriving data is essential for any life-long learning system. However, standard deep neural networks forget the knowledge about the old tasks, a phenomenon called catastrophic forgetting, when trained on incrementally arriving data. We discuss the biases in current Generative Adversarial Networks (GAN) based approaches that learn the classifier by knowledge distillation from previously trained classifiers. These biases cause the trained classifier to perform poorly. We propose an approach to remove these biases by distilling knowledge from the classifier of AC-GAN. Experiments on MNIST and CIFAR10 show that this method is comparable to current state of the art rehearsal based approaches. The code for this paper is available at https://bit.ly/incremental-learning