An Adaptive Fusion-Based Ensemble Deep Learning Framework for Cross-Subject EEG Emotion Recognition

1. A soft, attention-weighted electrode selection mechanism that replaces discrete hard selection with differentiable, correlation-based gating.

2. A multi-branch ensemble that fuses complementary spatial, topological, and temporal EEG representations.

3. A validation-driven adaptive weighting scheme that emphasizes stronger branches during the fusion process.

4. Evaluation on the SEED-IV dataset using a strict, subject-disjoint leave-one-subject-out protocol to address the cross-subject generalization gap.