In an interview, U.S. Intelligence Advanced Research Projects Activity (IARPA) program manager Jacob Vogelstein discusses the possibility of his agency saving money by refining existing algorithms that have been previously rejected for deep research using more rigorous training.
He notes many state-of-the-art systems need an interactive simulation environment to do more interesting artificial intelligence (AI) work. "With...continuous interactive feedback, you're going to have a machine-learning approach to control a robot, as opposed to programming it with a set of rules," Vogelstein notes.
He says IARPA has developed a large number of algorithms that have "laid in the shadows" for years because people saw them as having little utility for interesting AI tasks. Vogelstein notes this view has changed with the advent of scientific literature proving old algorithms can help meet challenges previously considered intractable when matched with more data and bigger computers.
"It's interesting to consider whether the money is better spent on the math of building new algorithms or really on assembling better datasets and putting together large computing resources to exploit those datasets using existing algorithms," he says.
Among the IARPA-funded AI projects Vogelstein cites are the Janus facial-recognition program, and the Machine Intelligence from Cortical Networks (MICrONS) program to explore brain-derived algorithms.
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA