Boundary-aware Prototype-driven Adversarial Alignment for Cross-Corpus EEG Emotion Recognition

Abstract

Electroencephalography (EEG)-based emotion recognition suffers from severe performance degradation when models are transferred across heterogeneous datasets due to physiological variability, experimental paradigm differences, and device inconsistencies. Existing domain adversarial methods primarily enforce global marginal alignment and often overlook class-conditional mismatch and decision boundary distortion, limiting cross-corpus generalization. In this work, we propose a unified Prototype-driven Adversarial Alignment (PAA) framework for cross-corpus EEG emotion recognition. The framework is progressively instantiated in three configurations: PAA-L, which performs prototype-guided local class-conditional alignment; PAA-C, which further incorporates contrastive semantic regularization to enhance intra-class compactness and inter-class separability; and PAA-M, the full boundary-aware configuration that integrates dual relation-aware classifiers within a three-stage adversarial optimization scheme to explicitly refine controversial samples near decision boundaries. By combining prototype-guided subdomain alignment, contrastive discriminative enhancement, and boundary-aware aggregation within a coherent adversarial architecture, the proposed framework reformulates emotion recognition as a relation-driven representation learning problem, reducing sensitivity to label noise and improving cross-domain stability. Extensive experiments on SEED, SEED-IV, and SEED-V demonstrate state-of-the-art performance under four cross-corpus evaluation protocols, with average improvements of 6.72\%, 5.59\%, 6.69\%, and 4.83\%, respectively. Furthermore, the proposed framework generalizes effectively to clinical depression identification scenarios, validating its robustness in real-world heterogeneous settings. The source code is available at \textit{https://github.com/WuCB-BCI/PAA}

Figures (7)

• Figure 1: Evolution of cross-domain feature alignment optimization. (a) Traditional global alignment method. (b) Prototype-guided local class-conditional alignment. (c) Contrastive semantic regularization, where intra-class and inter-class optimization are in opposite directions. (d) Maximize boundary-aware aggregation, dual-classifiers are used to identify and aggregate controversy samples.
• Figure 2: Unified PAA framework. Here, prototype learning is implemented by Eq. \ref{['Eq:proto']}; the Relation-aware Learning is implemented by Eq. \ref{['Eq:sim']}. The PAA framework instantiates three progressive configurations: PAA-L performs prototype-guided local class-conditional alignment; PAA-C encourages cross-domain intra-class compactness and cross-domain inter-class separability by contrasting semantic regularization; PAA-M is a fully boundary-aware configuration that integrates the maximize boundary-aware aggregation in the three-stage adversarial optimization scheme.
• Figure 3: The training process of PAA-M. (a) Firstly, PAA-M learns the base representation. (b) Then, ambiguous regions were identified by maximizing discrepancy, and samples from these regions achieved different results on the two classifiers. (c) Finally, the controversial samples are aggregated.
• Figure 4: Confusion matrices of baseline model PR-PL and proposed PAA-L, PAA-C, and PAA-M. The horizontal axis represents the predicted labels, while the vertical axis represents the true labels.
• Figure 5: T-SNE visualization of baseline model PR-PL and proposed PAA-L, PAA-C, and PAA-M.