clarke@utcsri.UUCP (05/25/87)
(SF = Sandford Fleming Building, 10 King's College Road) (GB = Galbraith Building, 35 St. George Street) SUMMARY: SYSTEMS SEMINAR, Tuesday, June 2, 11 am, SF1105 -- Yair Wand: "A System Theory Model for Formalizing Information Systems Design" COLLOQUIUM, Tuesday, June 2, 11 am, SF1101 -- David M. Regan: "Recovery and unconfounding of visual by human brain neurons" A.I. SEMINAR, Tuesday, June 2, 3 pm, SF1101 -- Eric L. Grimson: "Model Based Object Recognition and Localization" COMBINATORICS SEMINAR, Thursday, June 4, 3 pm, GB120 -- J. Fonlupt: "POLYNOMIAL ALGORITHM for solving the traveling salesman problem on certain classes of graphs" --------------- SYSTEMS SEMINAR, Tuesday, June 2, 11 am, SF1105 Professor Yair Wand University of British Columbia ``A System Theory Model for Formalizing Information Systems Design" Methodologies and techniques for systems analysis and design are based on practice and experience rather than being anchored in a theory. This work is based on the premise that a theoretical foundation for systems design can be found in general systems theory using ontological concepts. The idea is that an information system is a representation of the real sys- tem. The process of analysis and design is a transformation from user per- ceptions into a working information system. There must be some invariants of the transformation that will capture the semantics of the real sys- tem for the information system to be a 'good' representation. The identification of these invariants can be the basis for the analysis and design process. COLLOQUIUM, Tuesday, June 2, 11 am, SF1101 Professor David M. Regan University of Toronto and York University ``Recovery and unconfounding of visual by human brain neurons" In order to achieve precise eye-limb coordination, the brain must solve the problem of recovering the 3-D structive and 3-D motion of objects from the 2-D retinal image, and must unconfound different visual parameters. But any given neuron in primary visual cortex responds to several visual param- eters. I will describe psychophysical evidence that the human brain con- tains hardwired elements that use strongly nonlinear operations to recover and unconfound features of the external world, and that this process can more easily be understood in terms of achieving visually - guided motor action than in terms of extracting a veridical representation of the exter- nal world. Following these psychophysical findings, neurons with corresponding properties were found in the visual pathways of animals. There is evidence that spatial aspects of images are analysed by strongly nonlinear operations that allow the 25 sec arc sampling of the retinal image to be transcended, while unconfounding visual parameters. A new method for experimentally characterizing nonlinearities of human image pro- cessing will be described, providing a means of sharply distinguishing between nonlinear neural models of form and motion processing. A.I. SEMINAR, Tuesday, June 2, 3 pm, SF1101 Professor Eric L. Grimson M.I.T. ``Model Based Object Recognition and Localization" For the past several years, Tomas Lozano-Perez and I have been developing a framework for model-based object recognition from sensory data. The method is intended to be applicable to a wide range of sensing modalities, and assumes that the data is noisy, possibly sparse, and that the objects being sensed can be heavily occluded. The key to the method is the development of simple constraints on the relative shapes of the object models that can be used to rapidly reduce the search space of possible interpretations of the data. The method has been successfully applied to interpretation of tactile, laser range, sonar and visual data. Further- more, it can be extended to deal with parameterized families of objects, and can be amplified with a technique for automatically predicting optimal positions for obtaining additional sensory data. Examples of all of these areas will be presented. COMBINATORICS SEMINAR, Thursday, June 4, 3 pm, GB120 Professor J. Fonlupt University of Grenoble and University of Toronto ``POLYNOMIAL ALGORITHM for solving the traveling salesman problem on certain classes of graphs" We study a certain class of graphs for which the traveling salesman problem can be solved by a polynomial algorithm. These graphs are the graphs which cannot be reduced by deletion or contraction of edges to a well specified graph on six vertices. This algorithm generalizes results obtained by D. Ratliff and A. Rosenthal, G. Cornuejols, J. Fonlupt and D. Naddef, and includes results concerning the traveling salesman perfect graphs introduced by J. Fonlupt and D. Nad- def. This work is a joint work with A. Nachef. -- Jim Clarke -- Dept. of Computer Science, Univ. of Toronto, Canada M5S 1A4 (416) 978-4058 {allegra,cornell,decvax,linus,utzoo}!utcsri!clarke