AI for Health

🇬🇧 Domino Data Lab session- Increasing Data Science efficiency with on-demand, GPU-accelerated Ray clusters

Nov 17 09:55 - 10:25

Présentée par


In this session, we outline the problem of model complexity outpacing Moore's law, and how we can tackle the exponential increase of compute needs with parallelisation and GPU acceleration. We deliver a brief introduction to Ray, talk about how it can be used in conjunction with GPU-accelerated back-end compute, and show a brief technical demo of hyperparameter optimisation for ResNet50 against a 4GB dataset via RayTune. At the end of the session, we discuss the results of using the outlined approach for solving an identification of chemicals challenge, which was presented to us by our client Bristol Myers Squibb.