Bidirectional Time-Frequency Pyramid Network for Enhanced Robust EEG Classification

2025-10-11

AI Summary

Existing EEG-based recognition models suffer from poor generalizability across paradigms and subjects due to dataset specificity and inter-subject variability. To address this, we propose BITE—a unified, end-to-end model that jointly models motor imagery (MI) and steady-state visual evoked potential (SSVEP) tasks for the first time. Methodologically, BITE introduces a Pyramid Time-Frequency Attention (PTFA) mechanism and a time-frequency-aligned Bidirectional Temporal Convolutional Network (BiTCN). Leveraging STFT-based time-frequency alignment, multi-scale feature enhancement, and learnable fusion, it simultaneously captures dynamic time-frequency representations and amplifies discriminative neural patterns. Evaluated on four heterogeneous paradigm datasets, BITE achieves state-of-the-art performance in both intra-subject accuracy and cross-subject/cross-paradigm generalization, while maintaining high computational efficiency and broad applicability.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Multi-modal Vision
• Machine Learning: Mixture of Experts (MoE)

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Existing EEG recognition models suffer from poor cross-paradigm generalization due to dataset-specific constraints and individual variability. To overcome these limitations, we propose BITE (Bidirectional Time-Freq Pyramid Network), an end-to-end unified architecture featuring robust multistream synergy, pyramid time-frequency attention (PTFA), and bidirectional adaptive convolutions. The framework uniquely integrates: 1) Aligned time-frequency streams maintaining temporal synchronization with STFT for bidirectional modeling, 2) PTFA-based multi-scale feature enhancement amplifying critical neural patterns, 3) BiTCN with learnable fusion capturing forward/backward neural dynamics. Demonstrating enhanced robustness, BITE achieves state-of-the-art performance across four divergent paradigms (BCICIV-2A/2B, HGD, SD-SSVEP), excelling in both within-subject accuracy and cross-subject generalization. As a unified architecture, it combines robust performance across both MI and SSVEP tasks with exceptional computational efficiency. Our work validates that paradigm-aligned spectral-temporal processing is essential for reliable BCI systems. Just as its name suggests, BITE "takes a bite out of EEG." The source code is available at https://github.com/cindy-hong/BiteEEG.

Problem

• Addresses poor cross-paradigm generalization in EEG recognition
• Enhances robustness against individual variability in EEG classification
• Improves neural pattern capture through bidirectional time-frequency modeling

Innovation

• Bidirectional time-frequency pyramid network architecture
• Pyramid time-frequency attention for multi-scale enhancement
• Bidirectional adaptive convolutions capturing neural dynamics