Abstract: Machine learning in the real world comes with a slew of operational challenges. How do we monitor model training? How do we make records of how our data was preprocessed? How do we know what the predictions from that model actually look like? To answer these seemingly simple questions, many practitioners have had to build and maintain custom software systems. At Weights & Biases we are building a standard kit of developer tools for ML practitioners, making it easy to track and version your models and datasets, optimize model hyperparameters, visualize model performance and predictions, and more. In this session I will be doing a live demonstration of how a few lines of code can make your machine learning workflows dramatically more observable, reproducible, and scalable.
Bio: Ben is a machine learning solutions consultant with W&B. He trains our customers to use W&B and works with them to improve their machine learning workflow. Prior to joining W&B he was training models and developing ml infrastructure for Samsung Research.