Science Faculty Spotlight: Susan L. Epstein

December 14, 2020

Professor Susan L. Epstein has always been fascinated by how people and machines solve problems. At Hunter College and the Graduate Center, she works in artificial intelligence (AI), in particular in knowledge representation (how to describe the world to a machine) and machine learning (how a machine can be transformed by its experience). 

Trained first as a theoretical mathematician by Neal McCoy at Smith College and then Wilhelm Magnus at NYU's Courant Institute, Dr. Epstein went on to earn her Ph.D. in computer science at Rutgers The State University. Under the guidance of Natesa Sridharan, her thesis explored how a computer could do original mathematical research in graph theory. She spent her first few years at CUNY building the Graph Theorist, the first program to hypothesize and then prove fundamental theorems in graph theory. That also laid the groundwork for what she has done ever since —  develop knowledge representations and machine learning algorithms to support programs that learn to be experts. 

Interdisciplinarity is key in Dr. Epstein's research. She has collaborated with and learned from scientists in many disciplines: microbiology, robotics, linguistics, psychology, geography, mathematics, and urban planning. With them, she seeks important principles about knowledge and learning, and then helps computers exploit those ideas. Her current research interests include autonomous robot navigation and spatial cognition, but she also has long-standing interests in game playing, constraint satisfaction, intelligent drug design, and human-computer dialogue. A co-Principal Investigator at the National Science Foundation's Center for Brains, Minds, and Machines, she is also actively engaged in work on computational cognitive neuroscience. Funded primarily by the National Science Foundation, her research has produced 145 peer-reviewed papers. Journalists have written about her work, and she is often invited to speak about it, in interviews with the Wall Street Journal, CBS, NPR, and CNN.

Much of Dr. Epstein's work has involved the development of FORR (FOr the Right Reasons), a general architecture for learning and problem solving. FORR learns to revise its behavior, so that it gradually develops expertise at a set of problems. To make decisions, FORR integrates correct behaviors with large collections of heuristics and a variety of planners. FORR began as a system shell for Hoyle, the first program that learned to play 19 board games as well or better than the best human experts. Since then, FORR has grown in complexity and prowess as it has been applied to many other problem areas, including constraint satisfaction (ACE), urban park design (FLO), human-computer dialogue (FORRSooth), and robot navigation (SemaFORR). With its own Wikipedia page, FORR continues to inspire research in new problem areas.  

Dr. Epstein considers teaching a privilege, an opportunity to change forever how students experience the world. She won the first TIAA-CREF outstanding faculty award, and Hunter’s award for outstanding undergraduate mentors in science (for which one must be nominated by one’s students). She has supervised many graduate students and served on advisory committees for many more, both at CUNY and throughout the world. Her Problem Solving and Machine Learning laboratory challenges both undergraduate and graduate students with rigorous empirical research. Her mentees often receive prestigious awards, fellowships for graduate studies, and significant positions in academia and industry.

In service to the research community, Dr. Epstein is now an Executive Councilor for AAAI, the worldwide organization for artificial intelligence. She served for years on the governing board and then as the Chair of The Cognitive Science Society. She also was recently an officer of the Special Interest Group on Artificial Intelligence for ACM, the world’s largest organization for computer professionals. Meanwhile Dr. Epstein continues to be excited by how brains and minds solve problems, and how a computer can capitalize on that knowledge.