## Continual Learning and EEG-based Emotion Classification: A Critical Analysis Generalization to unseen subjects in EEG-based emotion classification remains a challenge due to high inter- and intra-subject variability. Continual learning (CL) offers a promising solution by learning from a sequence of tasks while mitigating catastrophic forgetting. ## Limitations of Regularization Methods A recent study examined the performance of regularization-based CL methods, such as Elastic Weight Consolidation (EWC), Synaptic Intelligence (SI), and Memory Aware Synapses (MAS), commonly used as baselines in EEG-based CL studies. The results indicate that these methods show limited performance for EEG-based emotion classification on the DREAMER and SEED datasets. Researchers identified a fundamental misalignment in the stability-plasticity trade-off. Regularization-based methods tend to prioritize mitigating catastrophic forgetting (backward transfer) over adapting to new subjects (forward transfer). In subject-incremental sequences, it was observed that the heuristics for estimating parameter importance become less reliable with noisy data and covariate shift. Furthermore, gradients on parameters deemed important often interfere with gradient updates required for new subjects, moving optimization away from the minimum. Importance values accumulated across tasks over-constrain the model, and performance is sensitive to subject order. Forward transfer showed no statistically significant improvement over sequential fine-tuning. ## Conclusions The high variability of EEG signals implies that past subjects provide limited value to future subjects. Consequently, regularization-based continual learning approaches are limited for robust generalization to unseen subjects in EEG-based emotion classification. This challenges the effectiveness of such methods in real-world contexts where data variability is high.