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

2026-04-01 · Source: cs.LG updates on arXiv.org · Field: Science & Research — Life Sciences & Biology, Health & Medical Research, Mathematics & Computational Sciences · Depth: Expert, extended

Summary

A new Prototype-driven Adversarial Alignment (PAA) framework has been developed to address performance degradation in cross-corpus Electroencephalography (EEG)-based emotion recognition. This framework tackles challenges like physiological variability, experimental paradigm differences, and device inconsistencies that cause models to fail when transferred across heterogeneous datasets. PAA is introduced in three configurations: PAA-L for prototype-guided local class-conditional alignment, PAA-C which adds contrastive semantic regularization for enhanced intra-class compactness and inter-class separability, and PAA-M, the full boundary-aware configuration. PAA-M integrates dual relation-aware classifiers within a three-stage adversarial optimization scheme to refine controversial samples near decision boundaries. Extensive experiments on SEED, SEED-IV, and SEED-V datasets demonstrate state-of-the-art performance under four cross-corpus evaluation protocols, showing average accuracy improvements of 6.72%, 5.59%, 6.69%, and 4.83% respectively. The framework also generalizes effectively to clinical depression identification.

Key takeaway

For AI Scientists and Machine Learning Engineers working on EEG-based emotion recognition across diverse datasets, adopting the PAA framework can significantly improve model generalization and robustness. Its progressive configurations, particularly PAA-M, offer a structured approach to mitigate cross-corpus distribution shifts and label noise. You should consider integrating prototype-driven alignment and boundary-aware adversarial optimization to achieve more stable and accurate emotion classification in real-world heterogeneous settings, including clinical applications like depression identification.

Key insights

The PAA framework improves cross-corpus EEG emotion recognition by aligning class-conditional distributions and refining decision boundaries.

Principles

• Global alignment alone is insufficient for cross-corpus EEG.
• Prototype-guided alignment preserves semantic structure.
• Dual classifiers can expose and refine boundary-sensitive samples.

Method

PAA uses a three-stage adversarial optimization: representation learning, discrepancy maximization with dual classifiers to expose boundary-sensitive samples, and boundary refinement to move samples away from ambiguous regions.

In practice

• Apply prototype learning for semantic structure preservation.
• Use contrastive regularization for better class separation.
• Employ dual classifiers to identify and refine decision boundaries.

Topics

• EEG Emotion Recognition
• Cross-Corpus Domain Adaptation
• Prototype Learning
• Adversarial Alignment
• Decision Boundary Refinement

Code references

• WuCB-BCI/PAA

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• UF-AMA: A unified framework for cross-domain emotion recognition via adaptive multimodal alignment — UF-AMA unifies multimodal physiological signal fusion with adaptive alignment and multi-level domain adaptation for robust cross-domain emotion recognition.
• Task-guided cross-subject latent alignment: a multi-encoder-decoder VAE — MED-VAE aligns neural activity across subjects using a pretrained ANN scaffold, eliminating the need for shared stimuli.
• Joint-Centric Dual Contrastive Alignment with Structure-Preserving and Information-Balanced Regularization — HILBERT aligns audio-text representations for long sequences by balancing modality contributions and preserving structure.
• How to Fine-Tune an SLM for Emotion Recognition — Fine-tuning SLMs for fine-grained emotion recognition on imbalanced datasets requires a multi-technique approach to data balancing and loss weighting.
• EEGDancer: Dynamic Emotion Latent Space Masked Modeling with Reinforcement Learning for EEG Continuous Emotion Prediction — Continuous EEG emotion prediction benefits from a unified framework combining discrete latent space learning, masked temporal modeling, and reinforcement learning.
• PsyScore: A Psychometrically-Aware Framework for Trait-Adaptive Essay Scoring and ZPD-Scaffolded Feedback — PsyScore unifies essay scoring and adaptive feedback using a shared psychometric latent ability representation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.