areas of interest Artificial intelligence, analogy, visual representations and reasoning, imitation, case-based reasoning, consciousness, theories of mind

advisor Dr. Ashok Goel research group Design Intelligence Laboratory

papers, patents, presentations, publications, and workshops 2010 2009 2008 1992 1987 1982 2010 Workshop on Visual Representations and Reasoning Do, E., Goel, A., Kunda, M., and McGreggor, K. proposed workshop for Twenty-Fourth AAAI Conference on Artificial Intelligence, July 2010 The last few decades have witnessed a lively debate on whether visual mental representations are a real part of human cognition. AI systems have likewise recruited visual systems of knowledge representation to varying extents, from proposition-based production systems to explicit models of visual knowledge. We propose this workshop on visual representations and reasoning to explore the nature and potential of visual thinking in intelligent agents of all kinds. The Interplay of Context and Emotion for Non-Anthropomorphic Robots McGreggor, K., Beer, J., Wiltgen, B., Jiang, K., and Thomaz, A. submitted to 5th ACM/IEEE International Conference on Human-Robot Interaction, March 2-5 2010, Osaka, Japan Household robots are becoming commonplace. The application of social cues, such as emotion, has the potential to make such robots easier to use and understand. However, it remains unclear how household robots can or should display emotion, and what considerations should be given to emotive behavior regarding the expected set of contexts in which the robot will operate. In this paper, we report the results of our systematic evaluation of the role both context and emotions play in the interpretation of context and emotion recognition of a non- anthropomorphic robot, the iRobot Roomba. Considerations, implications, and future work are discussed. A Fractal Approach towards Visual Analogy McGreggor, K., Kunda, M., and Goel, A. First International Conference on Computational Creativity, January 2010, Libson. We present a model of visual analogy that uses fractal image representations which rely only on the grayscale pixel values of input images, and are mathematical abstractions quite rigorously grounded in the theory of fractal image compression. We have applied this model of visual analogy to problems from the RavenÕs Progressive Matrices intelligence test, and we describe in detail the fractal solution strategy as well as some preliminary results. Finally, we discuss the implications of using these fractal representations for analogical reasoning and memory recall. 2009 Addressing the Raven's Progressive Matrices Test of "General" Intelligence Kunda, M., McGreggor, K., and Goel, A. Fall AAAI Symposium on Multimodal Representations, November 2009, Arlington, VA. The RavenÕs Progressive Matrices (RPM) test is a commonly used test of general human intelligence. The RPM is somewhat unique as a general intelligence test in that it focuses on visual problem solving, and in particular, on visual similarity and analogy. We are developing a small set of methods for problem solving in the RPM which use propositional, imagistic, and multimodal representations, respectively, to investigate how different representations can contribute to visual problem solving and how the effects of their use might emerge in behavior. Developing An Emotional Repertoire for Non-Humanoid Non-Verbal Robots Beer, J., Jiang, K., McGreggor, K., and Wiltgen, B. GT CoC Internal Report, May 2009 Household robots are becoming a commonplace. The application of social cues, such as emotion, has the potential to make such household robots easier to use and understand. However, it remains unknown how non-humanoid household robots, such as the iRobot Roomba vacuum cleaners, can display emotion, and in what contexts would emotion be most appropriate. We report a systematic evaluation of (1) investigating easily recognizable emotive behaviors of the Roomba, and (2) identifying contexts (edge conditions) in which these emotions may be appropriately applied. With this knowledge we developed a sophisticated emotional intelligence system which allows the Roomba to perceive, model and reason about itself and the world. The emotional intelligence system is experimentally evaluated by direct comparison to a standard non-emotive Roomba. Results of the two experiments suggest that the non-humanoid Roomba robot is capable of demonstrating emotive behaviors, and there are a number of edge conditions in which a person would expect the Roomba to act emotively. However, the application of an emotional intelligence system is a complex problem, and the authors failed to find user acceptance of the emotive robot. Considerations, implications and future work are discussed. 2008 Imitation as Analogy McGreggor, K. GT CoC Internal Report, December 2008 In this paper, I propose a novel cognitive account of imitation as a form of analogical reasoning. I build the case by examining what imitation is, what analogical reasoning is, and the relationship between them. I then illustrate an analogical architecture for imitation, and show how this view of imitation leads to a satisfactory solution of the correspondence problem. Finally, I briefly outline a new system and framework, I·Me·You, for conducting analogical experiments in the imitation context. 1992 Operating System Software Architecture and Methods for supporting Color Processing #5528261 Lindsay B. Holt, James A. Quarato, Jerry G. Harris, Ryoji Watanabe, Keith McGreggor Filed June 1992; Issued June 1996 An operating system software architecture, implemented in an object-oriented design, supports and processes color. The object-oriented design has two levels, one being a class (TColor) defining a virtual abstract base class and being a container class containing calibrated colors, the other being a class (TDeviceColor) defining a virtual abstract base class and a container class containing uncalibrated colors. Several calibrated color classes including a class (TXYZColor) defining XYZ color space descend directly from class (TColor), several uncalibrated color classes including a class (TRGBColor) descending directly from class (TDeviceColor), a class (TColorGamut) storing color gamut information of peripheral devices that may interface with the architecture, and a class (TColorProflle) storing tonal reproduction curves of the peripheral devices provide data structures, together with method functions for various color processing. The architecture is extensible to add new color classes as new color models and devices are developed, to add new color matching algorithms as desired, allows users to work in color space of their choice, and provides for color matching amongst any peripheral devices interfacing with the architecture. Color Processing System #5963201 Keith McGreggor, Christopher M. Yerga, David Van Brink Filed May 1992; Issued October 1999 A system for processing color information is based on carrying with color data, an indicator of the color space in which the data is represented. In this manner, the system is enabled to process color data from a variety of sources independent of the color space of the sources, because it is able to respond to the color space of a particular color value perform the necessary transformations to operate within any other arbitrary color and color space combination. The system provides for manipulating or combining colors independent of the source and destination color spaces. Also, the system operates on a per color component basis in user selected working color space, independent of the color space of the input or destination devices. Color Matching Apparatus and Method #5909291 Robin D. Myers, Keith McGreggor, Robert Johnson, Konstantin Othmer Filed March 1992; Issued June 1999 A color matching system initializes a translator by storing profiles of source and destination color devices which include the coordinates in a calibrated color space of the colorants produced in the source and destination devices and a tonal reproduction curve for each device. Mixing equations and parameters are precomputed to be used in matching calculations of an image including color pixels to be matched. The color gamut is divided into mixing regions in which a given point can be produced by a mixture of two chromatic colorants and the achromatic colorants of the destination device. A technique for selecting a mixing region is included which is based on the slopes of vectors defining the colorants of the destination device, and the sample color to be produced. Colors that are out of the gamut of the destination device are adjusted according to precomputed parameters in order to preserve the lightness, chromaticity, or other selected characteristics of the sample color as suits the need of a particular application. The system is particularly suited for computer systems independent of the source and destination devices to be used, and can be adapted to run with a translation cache to enhance the speed of operation of translating a color image on one device for display on a second device. earlier works as an author or significant contributor 1987 Artificially Intelligent Robots McGreggor, K., and Troutman, C. In: The Southern Manufacturing Technology Conference Proceedings, January 1987, Charlotte, NC Developing artificially intelligent robots can be the basis for integrating AI techniques into existing and planned automated workcells. The described application makes use of case-based reasoning and natural language process techniques. 1982 Towards a Computer Model of Psychiatric Reasoning Kolodner, Janet L., and Kolodner, Robert M. Proc Annu Symp Comput Appl Med Care. 1982 November 2; 99Š103. Human experts are able to introspect about their knowledge and learn from past experience. It is this view of expertise that we are exploring. This paper will present the basis for this view, the reasoning model it implies, and a computer program which begins to implement the theory. The program, called SHRINK, simulates a psychiatrist. The role of experience in development of expertise Kolodner, J. In: Proceedings of the American Association for Artificial Intelligence (1982, August), pp. 273-277 Pittsburgh, PA . Perhaps the most distinguishing feature of an expert is that when given a novel problem to solve in his or her domain of expertise, the expert can solve the problem easily. Novices, on the other hand, are good at dealing with typical problems or "classic" cases, but not novel problems. In people, the evolution from novice to expert happens as a result of being able to introspect and examine the knowledge used in solving problems. That introspection and examination allows people to learn from experience. A human expert can interpret a new case in terms of something (either a previous case or generalized knowledge) he is already familiar with. This implies that as an expert is having new experiences, he is evaluating and understanding them in terms of past experiences. In the process, he is integrating the new experience into his memory so that it too will be accessible to use in understanding a later case. Intelligent Fact Retrieval Kolodner, Janet L., Simpson, B., Sharpe, D., and McGreggor, K. ACM SIGART Bulletin archive. Issue 79, January 1982 An important problem facing people working in the field of Natural Language Processing is specifying, representing, and organizing the knowledge necessary for understanding. As humans, we build up that knowledge from experience. But, how can we represent and organize the right kinds of knowledge in the computer? Furthermore, as new information gets added to the computer memory, how can we make the computer automatically integrate that knowledge into memory in smart ways so that it can be easily accessed and "remembered" when necessary for understanding, and without irrelevant knowledge getting in the way?