Graph-Based Learning of Spectro-Topographical EEG Representations with Gradient Alignment for Brain-Computer Interfaces

2025-12-08

AI Summary

To address low inter-class separability in EEG signals caused by substantial inter-subject variability and strong non-stationarity, this paper proposes a graph neural network-based multi-domain fusion framework. It constructs an EEG frequency-topological graph to model functional connectivity among channels and jointly learns cross-domain representations with time-frequency spectral features. A gradient alignment mechanism is introduced to mitigate optimization conflicts arising from multi-source feature fusion, while center loss and pairwise discrepancy loss are jointly employed to enhance discriminability. The method achieves significant classification performance improvements on the BCI-2a, CL-Drive, and CLARE datasets; ablation studies confirm the substantial contribution of each component. The core innovation lies in the synergistic integration of frequency-topological structural modeling with gradient alignment-driven multi-domain collaborative optimization, thereby substantially improving robust discriminative capability for cross-subject and cross-session EEG signals.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Learning & Optimization for CV
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

We present a novel graph-based learning of EEG representations with gradient alignment (GEEGA) that leverages multi-domain information to learn EEG representations for brain-computer interfaces. Our model leverages graph convolutional networks to fuse embeddings from frequency-based topographical maps and time-frequency spectrograms, capturing inter-domain relationships. GEEGA addresses the challenge of achieving high inter-class separability, which arises from the temporally dynamic and subject-sensitive nature of EEG signals by incorporating the center loss and pairwise difference loss. Additionally, GEEGA incorporates a gradient alignment strategy to resolve conflicts between gradients from different domains and the fused embeddings, ensuring that discrepancies, where gradients point in conflicting directions, are aligned toward a unified optimization direction. We validate the efficacy of our method through extensive experiments on three publicly available EEG datasets: BCI-2a, CL-Drive and CLARE. Comprehensive ablation studies further highlight the impact of various components of our model.

Problem

• Learns EEG representations for brain-computer interfaces
• Improves inter-class separability of dynamic EEG signals
• Aligns conflicting gradients from multi-domain EEG data

Innovation

• Graph convolutional networks fuse EEG frequency and time-frequency embeddings
• Center and pairwise loss enhance inter-class separability for EEG
• Gradient alignment resolves domain conflicts for unified optimization