Delivering business value from a machine learning project requires practitioners to iterate on vast amounts of data as well as execute many different experiments. By the time a given project is producing production-ready results, the relationship between data used and computation executed can become incredibly difficult to keep organized.
At the end of a given project, how does a machine learning team effectively communicate their findings to other parts of the organization? How would a team lead easily understand if the current project efforts are trending in the right direction, when the work is dispersed across a number of team members (or separate teams)? How would an executive ensure their organization’s machine learning efforts are documented in a centralized fashion for auditability and risk management?
In this talk we’ll explore three principles of an ideal machine learning workflow: traceability, reproducibility, and collaboration, and see how the Weights & Biases platform can easily enable organizations to achieve these at scale.
Jack Bailin is the solutions engineering lead at Weights & Biases. His focus is on helping enterprise AI machine learning teams adopt best practices for scalable workflows, from code instrumentation with MLOps platforms/tooling all the way through effective dissemination of information across the organization.
Jack got his start in machine learning during his undergraduate research studies in Physics, when he implemented a deep learning classifier for identification of basal cell carcinoma skin tissue based on spectral data input. He then spent time as the lead machine learning engineer at Foyer AI, where he built and productionalized deep learning models for classification and semantic segmentation of real estate photography.