April 2022 Summaries

A StatQuest tutorial introduces PyTorch through a step-by-step neural network example, covering how to build a network from scratch, generate and graph predictions, and optimize a parameter using backpropagation. It also offers a downloadable Jupyter Notebook containing the code demonstrated in the accompanying video.

University of British Columbia economist Jesse Perla and his team use machine learning to address computationally demanding macroeconomic research, including deep-learning solutions for high-dimensional models with PyTorch Lightning and Bayesian optimization in Julia for faster model estimation. Previously, infrastructure complexity, limited local computing capacity, and reliance on institutional clusters or personal laptops slowed experimentation and required expertise beyond the team’s core research focus. Grid.ai enabled the researchers to run the same code locally and in the cloud without learning DevOps tools such as AWS, Kubernetes, or Linux, allowing them to launch scalable training jobs directly from GitHub. The platform’s versioning, reproducibility, shared dashboard, multi-language support, and collaboration features helped students and researchers manage runs, regenerate results, share work, and conduct larger hyperparameter sweeps more efficiently.