Generative Negative Replay for Continual Learning
This addresses catastrophic forgetting in continual learning for AI systems, offering a novel replay strategy that is effective in challenging, real-world scenarios, though it is incremental as it builds on existing generative replay methods.
The paper tackles catastrophic forgetting in continual learning by proposing generative negative replay, where generated data from old classes serve as negative examples to improve learning of new classes, especially with small training experiences. The approach is validated on complex, high-dimensional datasets like CORe50 and ImageNet-1000, where existing generative replay methods typically fail.
Learning continually is a key aspect of intelligence and a necessary ability to solve many real-life problems. One of the most effective strategies to control catastrophic forgetting, the Achilles' heel of continual learning, is storing part of the old data and replaying them interleaved with new experiences (also known as the replay approach). Generative replay, which is using generative models to provide replay patterns on demand, is particularly intriguing, however, it was shown to be effective mainly under simplified assumptions, such as simple scenarios and low-dimensional data. In this paper, we show that, while the generated data are usually not able to improve the classification accuracy for the old classes, they can be effective as negative examples (or antagonists) to better learn the new classes, especially when the learning experiences are small and contain examples of just one or few classes. The proposed approach is validated on complex class-incremental and data-incremental continual learning scenarios (CORe50 and ImageNet-1000) composed of high-dimensional data and a large number of training experiences: a setup where existing generative replay approaches usually fail.