On today’s program, some kickoff analysis of the Nvidia/Arm deal; a look at geospatial AI and what it requires in terms of future hardware and frameworks; physics-informed AI and what it means for future simulations and use cases; the Summit supercomputer and the “right” architecture for large-scale, multidisciplinary COVID-19 research; the elements of a hyperconverged data platform; infrastructure as code; much more in today’s interview lineup. More info and timestamps below.
We have a wide-ranging program today, kicking off with an analysis of the geospatial industry and its growing adoption of deep learning to speed products to market. The challenge is that existing AI frameworks aren’t suited to data types and requirements and the hardware will also take some changes over time with San Gunawardana, co-founder and CEO of Enview.
Also on today’s show we talk with Christopher Lamb, VP of Computing Software at Nvidia about physics-informed neural networks and where these are taking hold. We talk about where this approach might shave off training times and model data for a wide range of potential use cases.
We also talk with Dan Jacobson, lead researcher and computational systems biologist at Oak Ridge National Laboratory about base of research targeting COVID-19 on the Summit supercomputer. We look ahead to future architectures and project what they might be able to lend to similar research at exascale in the future.
2:23 – Geospatial AI: The Current State and Future System, Framework Requirements
14:45 – Infrastructure as Code: Filling in the Missing Gaps
24:19 – Physics-Informed AI: What it Means for Training Times/Efficiency, Use Cases
34:23 – Summit of COVID-19 Research: Large-Scale Modeling and Future Architectures
43:53 – Piecing Together the Elements of a Hyperconverged Data Platform