Domain Incremental Learning for EEG-Based Seizure Prediction

When building seizure prediction systems, the typical research scenario is patient-specific. In this scenario, the model is limited to performing well for individual patients and cannot acquire knowledge transferable to new patients to learn a set of universal parameters applicable to all patients. To this end, we investigate a new task scenario, domain incremental (DI) learning, which aims to build a unified epilepsy prediction system that performs well across patients by incrementally learning new patients. However, the neural network is susceptible to the problem of catastrophic forgetting (CF) during incremental training, which quickly forgets the knowledge learned from past tasks due to differences in domain distributions. To address this problem, we introduce an experience replay (ER) method, which stores a few samples from previous patients and then replays them in new patient training to review past knowledge. In addition, we propose a novel ER-based centroid matching method (ER-CM) that computes the class centroid in the feature space using subsets stored in the memory buffer. The ER-CM regularizes incremental training by matching the distance between sample embeddings and class centroid, providing additional guidance for parameter updates. Experimental results demonstrate that the ER approach substantially reduces CF and significantly improves performance when combined with CM.