Lightweight Adaptation of EEG Foundation Models for Stroke Motor Imagery Decoding: Domain Shift and Subject-Level Robustness
The episode discusses a paper on adapting foundation models for decoding motor imagery in stroke patients. Hosts explore how relying solely on healthy subject benchmarks is insufficient due to domain shift, where performance drops significantly in real-world stroke data. The hosts conclude that using lightweight adaptation methods like LoRA enables efficient, personalized AI that can be scaled for clinical use.