Altaheri/EEG-ATCNet

Attention temporal convolutional network for EEG-based motor imagery classification

What it solves

This project implements a neural network architecture designed to classify motor imagery (MI) EEG signals. It addresses the challenge of accurately decoding brain signals associated with imagining movement to enable better brain-computer interfaces (BCI).

How it works

ATCNet uses a physics-informed approach combining three main components:

  1. Convolutional (CV) block: Extracts low-level spatio-temporal information from EEG signals into high-level temporal representations.
  2. Attention (AT) block: Uses multi-head self-attention (MHA) to highlight the most critical information within the temporal sequence.
  3. Temporal Convolutional (TC) block: Extracts high-level temporal features from the highlighted data.

The model also employs a convolutional-based sliding window to augment data and improve classification performance. It supports various attention schemes including multi-head self-attention (mha), locality self-attention (mhla), squeeze-and-excitation (se), and convolutional block attention modules (cbam).

Who it’s for

Researchers and developers working on EEG-based motor imagery classification, brain-computer interfaces, and neural signal processing using TensorFlow.

Highlights

  • Physics-informed architecture: Combines convolutional layers with attention mechanisms and temporal convolutional networks.
  • Benches against multiple models: Includes implementations of EEGNet, EEG-TCNet, DeepConvNet, and others for comparison.
  • Flexible attention: Supports multiple interchangeable attention mechanisms (MHA, MHLA, SE, CBAM).
  • Evaluation strategies: Provides tools for both subject-specific and subject-independent (Leave One Subject Out) evaluation.

Related

  • Project
    githubharald/SimpleHTR

    A TensorFlow-based handwritten text recognition system that converts images of single words or text lines into digital text using CNN and LSTM layers.

  • Project
    CaviraOSS/OpenMemory

    A cognitive memory engine for AI agents that provides long-term, self-hosted memory with temporal reasoning and multi-sector categorization instead of simple RAG.

  • Dispatch
    The State of Computer Vision at Hugging Face

    Hugging Face has expanded its ecosystem to support 8 core computer vision tasks, over 3,000 models, and 100+ datasets, integrating both Transformer and convolutional architectures.

  • Project
    facebookresearch/mobile-vision

    A collection of hardware-aware efficient ConvNet models and tools, including FBNet and ChamNet, designed to optimize computer vision performance and latency on mobile devices.

  • Project
    philipperemy/keras-tcn

    A Keras implementation of Temporal Convolutional Networks (TCN) that provides a more stable and parallelizable alternative to LSTMs and GRUs for long sequence modeling.