Presentation #541.02 in the session “Computational Augmentation to Surveys and Science Programs”.
We describe our usage of the Nvidia Jetson Nano, a small embedded compute board with four CPU cores and a 128-core Maxwell generation GPU. The hardware is well suited for rapid Python prototyping and developing neural network applications. We utilize the PyTorch framework to train a convolutional neural network image classifier. The image data are a small subset of the VLA Sky Survey that is split into both testing and validation collections. We construct a simple shape classification scheme and report on the loss function for the network, all run on the Nano hardware.