Focal Interest
Focal Interest

Deep neural networks

  1. LPS-GNN scales graph learning to 100-billion-edge graphs
    LPS-GNN enables efficient graph neural network processing on 100 billion-edge graphs using single GPUs, achieving 13.8% improvements in user acquisition tasks through novel graph partitioning.
  2. Binarized MLP-Mixer trained successfully on MNIST
    Binarized MLP-Mixer achieves 0.9+ accuracy with quantized activations on MNIST, enabling efficient deployment on IoT edge devices without convolutional or attention mechanisms.