This $1.2 million project grant from the National Science Foundation's Computer and Information Science and Engineering program will fund research at the Massachusetts Institute of Technology to develop new techniques for interactive machine learning with rich feedback. Over a three-year period from September 2022 to August 2025, the grantee will pursue three objectives: establishing a framework for grounding complex feedback using simple supervisory signals; creating algorithms allowing agents to proactively solicit feedback; and developing tools to assist human supervisors in selecting maximally informative feedback. A subaward of $100,000 will support related research at the University of Washington aiming to improve artificial intelligence learning similar to natural human environments. The grantee intends to evaluate the research in simulated environments involving navigation, robotics, and assembly tasks, assessing benefits to sample efficiency, development time, and usability.
Generated 1/6/24, 10:42 PM