The sub-area of AI concerned with human languages (“natural languages”) is computational linguistics. Researchers in this area are interested in developing programs that can “understand” and generate natural language. “Understanding” involves parsing linguistic input, determining its literal and non-literal meaning and representing the meaning in a computational formalism; generation reverses this process. Research in this area is now being applied in commercial systems for tasks such as automatic or semi-automatic translation from one language to another, information retrieval and intelligent aids to writers.
A theme common in much AI research is that to behave intelligently, computers must come to “know” a good deal of what every human being knows about the world and the organisms that inhabit it. Knowledge representation and reasoning (or KR) is the study of how to impart this knowledge to a computer: how do we write down descriptions of the world in such a way that a computer would be able to draw appropriate conclusions about the world by manipulating them?
Once knowledge is represented effectively, action selection to create a behavioural strategy involves solving a variety of planning and constraint satisfaction problems, often in the face of uncertain information.
Here at the University of Toronto, the work of Reiter and Levesque most recently has been concerned with “cognitive robotics,” that is, KR from the point of view of an autonomous robot interacting with a dynamic and incompletely known world. Among other things, this has required developing new accounts of the relationship among the knowledge, perception and action of such a robot.
Our group invents and applies novel learning algorithms for machine learning, neural networks and statistical pattern recognition. Our tools include probability theory, Bayesian methods, graphical models, learning theory and mean field methods from physics. Our main aims are to develop synthetic intelligence algorithms that are useful for practical tasks, to understand the relationship of machine learning and neural network algorithms with other statistical approaches, and gain insight into how the brain performs supervised and unsupervised learning and how it represents information in the neural code.
The long-term goal of research in computational vision is to understand the visual information that is represented in images and image sequences. By “understanding,” we mean that a computer system viewing an image could report on the contents of an image in a useful manner, where utility may be measured by specific tasks or by the standards of human perception. Research in the field ranges from practical industrial vision applications to the design and construction of robotic vision sensors (such as stereo heads) to attempts to understand how the human brain processes and uses visual information. As a result, there are many sub-areas of research within computational vision, including edge detection, segmentation, texture analysis, colour perception, stereo tracking, perceptual organization, object recognition, active and attentive vision, sensor design, motion analysis, event perception, learning and so on. Impressive successes have been seen, but the research area contains a large number of open problems, making this an intriguing and challenging topic for many years to come.
Our group’s members have received many prestigious awards and grants including Premier’s Research Excellence Awards, NSF Career awards, Sloan Fellowships, Steacie Fellowships, NSERC University Faculty Award, ONR Young Investigator Award, Fellowships in the Royal Societies of Canada and Britain and in the American Association of Artificial Intelligence, two major awards from IJCAI, the David Marr Prize in Computational Vision, and several best-paper prizes at conferences. Many graduates of the group are now distinguished AI researchers themselves.
The AI group has also been recognized and supported by Canadian federal government grants such as NSERC and organizations such as IRIS, CITO, MITACS, Bell University Labs, Canadian Institute for Advanced Research, Xerox, and IBM.