The Defense Advanced Research Projects Agency has picked research teams that will develop new machine learning methods designed to enable artificial intelligence systems to adapt to new changes and implement previous knowledge and skills to new situations.
DARPA said Thursday teams from the University of California in Irvine, Tufts University and University of Wyoming will perform work on two technical areas of the agency’s Lifelong Learning Machines program.
The L2M program’s first technical area will deal with the development of complete platforms and components, while the other will look at learning methods in biological organisms and transition those mechanisms into computational processes.
The University of California team plans to develop a machine learning system designed to predict potential outcomes through comparison between inputs and existing memories.
The Tufts University team has begun to investigate how salamanders and other animals regenerate in an effort to develop robots designed to adjust to changing environments through alteration of function and structure.
The University of Wyoming team will develop a computational system designed to identify modular memories through the use of context and rearrange those memories with sensory input to establish behaviors in order to correspond to new circumstances.
“With the L2M program, we are not looking for incremental improvements in state-of-the-art AI and neural networks, but rather paradigm-changing approaches to machine learning that will enable systems to continuously improve based on experience,” said Hava Siegelmann, L2M program manager.
“L2M seeks to enable AI systems to learn from experience and become smarter, safer and more reliable than existing AI,” Siegelmann added.