Webinar
Accelerating Deep Learning for Molecular Modeling Applications
Learn how to achieve faster time to results and state-of-the-art performance for GNN model SchNet
In life sciences, machine learning has emerged as an attractive alternative to expensive computational methods for the prediction of chemical properties and generation of molecular structures. Most recently, the development of SchNet, a novel deep learning architecture for molecules and materials, has advanced the accuracy of modeling quantum interactions in molecules.
Improving the scalability of training these next generation machine learning models is essential for accelerating scientific discovery. Molecular datasets are uniquely structured, requiring hardware capable of processing millions of small graphs with widely varying size and sparsity characteristics. This complexity can pose a challenge for traditional processing technologies, limiting real-world deployment at scale.
In this webinar, the US Department of Energy’s Pacific Northwest National Laboratory and Graphcore will explain how novel IPU-based systems can achieve state-of-the-art performance for SchNet, driving faster time-to-results for molecular applications.
What you'll learn:
Meet the speaker
Ellie Dobson
VP Product Cloud, Graphcore
Ellie Dobson has 15 years' experience building data-driven solutions in engineering organisations. She is the VP Product, Cloud at Graphcore supporting the company's mission to drive new breakthroughs in machine intelligence with innovative AI systems and services.
Ellie is a particle physicist by training, having worked as part of the Higgs discovery team at CERN. She has worked for the past decade for software companies in variety of industries, but with the focus on the application of advanced analytics and digitalisation to engineering disciplines.
Meet the speaker
Sutanay Choudhury
Chief Scientist, Data Science
Pacific Northwest National Laboratory
Dr. Choudhury has 10+ years of experience in large-scale graph analytics and machine learning. His research focuses on learning high-fidelity representations of structure from complex data sources and development of methods for reasoning and prediction.
He has served as PI, co-PI, and contributor on multiple projects funded by the U.S. Department of Defense, U.S. Department Homeland Security, U.S. Department of Energy, and PNNL’s internal research programs focusing on designing extreme-scale semantic databases, scalable inference using probabilistic graphical models, data-driven construction of knowledge graphs from unstructured text, and simulators for synthetic generation of property graphs.