This week's colloquium: "A comparison of neural network frameworks" by Erik Spence from SciNet. The Compute Ontario Colloquia are weekly Zoom presentations on Advanced Research Computing, High Performance Computing, Research Data Management, and Research Software topics, delivered by staff from three Compute Ontario consortia (CAC, SciNet, SHARCNET) and guest speakers. The colloquia are one hour long and include time for questions. No registration is required. Most presentations are recorded and uploaded to the hosting consortium video channel.
The standard approach to programming neural networks is to use a neural network programming framework. Neural network frameworks are specifically designed to make implementing neural networks easy, and development fast. However, those beginning their journey in programming neural networks may be unfamiliar with the available frameworks, and the advantages of each. In this talk two solutions to the standard MNIST problem will be presented, using two commonly-used neural network frameworks: PyTorch Lightning and Keras/Tensorflow. We will compare the two frameworks for ease of implementation, performance, and training speed. The advantages and disadvantages of the two frameworks will be discussed. Familiarity with neural network and Python programming will be assumed.