Work Environment
Programming is primarily done using Python and the PyTorch machine learning framework. Since deep learning tasks require significant time and hardware resources, our professor provides several high-VRAM GPUs for model training, including P40, 4090, and 3090 desktop graphics cards with approximately 24GB of VRAM. This allows for larger batch sizes during training.
Job Description
Model training using the PyTorch framework combined with other various data processing technologies.
Role Played in Research
Jason Guo:
1. Public and private ECG database data collection preprocess, labeling, and normalization.
2. Transformer model for r peak detection and normal-abnormal classification.
Yang-I Lin:
1. Augmenting 2.5 second beat using unconditional and conditional Wasserstein GAN with Gradient Penalty (WGAN-GP).
2. Augmenting 10 second ECGs using Structured State Space Sequence Model (S4) with Diffusion Model.