Continual learning for seizure prediction via memory projection strategy

Yufei Shi, Shishi Tang, Yuxuan Li, Zhipeng He, Shengsheng Tang, Ruixuan Wang, Wei-Shi Zheng, Ziyi Chen, Yi Zhou

Abstract: Highlights•Proposed Patient-IL adapts to dynamic patient increase for seizure prediction, facilitating EEG-based incremental learning studies.•Proposed MP strategy tackles CF in epilepsy prediction due to EEG data interference, integrating smoothly with algorithms, exhibiting strong flexibility and portability.•Innovative algorithm constructs dynamic projection matrices, optimizing with weak correlation to reduce interference, promote knowledge transfer, and minimize computation via dimensionality reduction.•Extensive experiments & ana lyses prove MP effectively overcomes CF in Patient-IL. Results across diverse datasets show MP reduces forgetting rate <5%, matching top CL strategies.