Speaker: Paul Cohen , University of Massachusetts
It's about time: Challenges for Robot Baby
Abstract: Time is a theme that runs through all my work. Timing is important in wargaming and planning; simulations produce vast multivariate time series that require analysis; intelligence data becomes available over time and must be assembled into coherent episodes or stories. In our Robot Baby project, time series of sensor and perceptual data are the raw material for learning contentful representions. I will talk about algorithms for finding structure in temporal data, both univariate and multivariate, categorical and continuous. Specifically, I will describe algorithms for clustering similar time series, extracting hierarchical event descriptions from series, and segmenting series into episodes. Robot Baby will be my focus because it illustrates the range of issues that arise in my work. At the philosophical end of the range, I will discuss semantic autonomy, the idea that meanings of symbols are acquired by and for agents. Robot Baby, for example, learns the meanings of words in several languages. I will conclude with two challenge problems that have valuable applications and also drive research in the direction of semantic autonomy for agents.
Short Bio: Paul R. Cohen is a Professor in the Department of Computer Science at the University of Massachusetts, and Director of the Experimental Knowledge Systems Laboratory. His PhD is from Stanford University in Computer Science and Psychology, in 1983, and his MS and BA degrees in Psychology are from UCLA and UC San Diego, respectively. Dr. Cohen served as a Councillor of the American Association for Artificial Intelligence, 1991-1994, and was elected in 1993 as a Fellow of the AAAI.
Host: Robert St. Amant, Computer Science, NCSU