Cloud is key to scaling AI research
- By Matt Leonard
- Oct 24, 2018
Getting academic researchers the computing power they need for artificial intelligence is a work in progress, according to James Kurose, the assistant director of the National Science Foundation for Computer and Information Science and Engineering.
While federal agencies tend to have access to high-performance computing capabilities -- the national labs have some of the fastest supercomputers in the world -- computing resources at the higher education level tend to be more modest, Kurose said.
“We need to put computation in the hands of computer scientists,” he said on a panel at an Oct. 23 conference hosted by NVIDIA. “You can have great ideas, but if you can’t do them at scale, sometimes they just remain great ideas.”
Kurose predicted that working with cloud providers will help fill in the gap. NSF already has agreements with Amazon Web Services, Google, Microsoft and IBM where each company donates millions of dollars worth of cloud credits to NSF-funded researchers.
This is great for getting started, he said, but there must be “more systemic programs for getting cloud into the hands of NSF funded researchers.”
A better framework for public-private partnerships could help spread resources to where they’re needed, according to Howie Choset, a professor of robotics at Carnegie Mellon University.
Matt Leonard is a former reporter for GCN.