Artificial Intelligence

Artificial Intelligence is the study of computational models of the mind. At Yale, there are a wide variety of topics studied, including knowledge representation, inference, planning, learning, vision, and robotics.

The term ``artificial intelligence'' is somewhat misleading, because the focus of research in the field is often on more mundane activities, such as simple visual perception, than the word ``intelligence'' would suggest. The field has learned over the years that the effortlessness of a skill such as vision is deceptive, that in fact the brain does a great deal of hard labor behind the scenes to allow us to see without conscious effort. It will take us years to duplicate the skills that nature evolved over eons.

AI uses many of the same techniques as other areas of Computer Science application, from numerical optimization to symbolic indexing. The key to solving any problem is always the algorithm and its analysis. The goal is always to characterize precisely a set of problems and demonstrate an algorithm that solves them with reasonable efficiency. But, at least at its current state of development, AI is of necessity more exploratory than other areas. We are often forced to define a problem at the same time that we try to solve it. It often happens that we don't know how to analyze the performance of an algorithm with existing tools, but we believe that its average-case performance is much better than its worst-case performance, and this belief must be backed up with experiments. Sometimes a piece of research is valuable even though all that it accomplishes is the discovery of a dimension of knowledge representation that we as yet do not know how to incorporate into algorithms at all.

In general we think it is a mistake for AI research to focus on central mental function and ignore input and output. In the long run, machines will not be treated as intelligent unless they can perceive and manipulate the objects around them. Real perception and action impose stubborn constraints on thinking. Sophisticated robot planning is wasted if the robot crashes into the wall while trying to generate a predicate-calculus description of the world in front of it. So our planning research focuses on models of execution and replanning in realistic, changing worlds, and not so much on provably correct plans.

Here is a partial list of the projects we are now working on:

• Map learning for mobile robots: How can a robot get back to a place it's been before? To answer this question, you need to define what a place is. We are pursuing a model in which a place is a perceptually distinctive point, reachable from other places by fairly robust but not perfect operations. As the robot wanders around its world, it builds up a graph of places and the metrical relationships among them. It is able to correct errors in this graph by merging two places that are really the same, and splitting a place that really corresponds to two similar places.

• Perception and decision making: The information that an autonomous system has available is often incomplete and may contain errors. In order to carry out its activities, the system must use sensors to gather information. Of course, a central problem area is the development of methods for interpreting and using sensor information. However, these methods must match the capabilities of the system. Consequently research also focuses on optimizing the use of sensing equipment and computer resources to gather and process information for a particular task.

• Transformational reactive planning: A plan is a program an agent gives to itself in response to problems it encounters. In changing worlds, the plan must contain rules for reacting to problems, but in addition there must be a planning module with a more long-range outlook that can revise the plan before execution is completed. Our planner revises the plan by using information gained by mentally simulating, or ``projecting,'' it to see how execution will go.

• Engineering problem solving: Scientists and engineers model, analyze, design, and explain complicated physical systems using reasoning that is often heuristic, qualitative, and graphical in addition to formal mathematical arguments. Understanding this ability requires a deep understanding of the techniques in simplifying mathematical models, finding approximate solutions, and deriving consequences from mathematical models by mixed symbolic, numerical, and geometric computations. Current research involves the automatic analysis of certain classes of fluid dynamics problem. Our goal is to produce computer programs that not only serve as high-level assistants to scientists and engineers but also smart enough to carry out independent research in fluid dynamics and write research papers of publishable quality.

• Object Recognition: One of the fundamental goals of computer vision is to recognize the three-dimensional objects in a single perspective projection image in terms of a database of object models. Most of the research has considered polyhedral objects where a correspondence between object features (e.g. vertices) and image features (e.g. corners) can be readily established. At Yale, we have focussed on recognition of curved objects where it is more difficult establish such a correspondence. Specific problems that are being considered include representing objects by algebraic surfaces, building object models from a sequence of images, constructing aspect graphs (a viewer-centered representation enumerating all topologically distinct line drawings), estimating an object's pose from one image, and indexing a model database.

Faculty members working in Artificial Intelligence are Gregory Hager, David Kriegman, Drew McDermott, Francois Meyer, Willard Miranker, Marcello Pelillo, and Steven Zucker. Michael Hines, and Kaleem Siddiqi are Research Scientists.