Supercomputer Institute Vol. 12 No. 4 Behavioral Sci. Symp.

Symposium on Supercomputer Applications in the Behavioral Sciences held May 10-12, 199

This symposium was a follow-up to the 1985 National Science Foundation conference Advanced Computing for Psychology, which examined the increasing importance of high-performance computers, including workstation clusters, in the behavioral sciences. At the current conference, held Friday, May 10 through Sunday, May 12, 1996 at the Supercomputer Institute, representative topics included computer-intensive simulation methods, virtual reality, human factor, neural networks, brain magnetic resonance imaging, information processing, large data handling, perception and human vision, and graphic visualization in the behavioral sciences. This symposium was sponsored by the University of Minnesota Supercomputer Institute, the College of Education and Human Development, and the Federation of Behavioral, Cognitive, and Psychological Sciences. The organizing committee consisted of Lynne Edwards, Department of Educational Psychology and the Supercomputer Institute; Stephen Link, Federation of Behavioral, Cognitive, and Psychological Sciences; and Cynthia Null, NASA Ames Research Center.

In addition to talks, presentations, and discussions, the symposium included tours of high-performance computing facilities at the University of Minnesota. These included the IBMWorkstation Cluster, the Laboratory for Computational Science and Engineering, and the supercomputing facilities at the Minnesota Supercomputer Center Inc.

The presentations began on Saturday morning with welcoming speeches by Donald Truhlar, Director of the Supercomputer Institute and Department of Chemistry, followed by Cynthia Null, NASA Ames Research Center, who was also a co-organizer for the 1985 Conference on Advanced Computing for Psychology, and Stephen Link, Federation of Behavioral, Cognitive and Psychological Sciences, who participated in the 1985 NSF conference. Richard Shiffrin and Peter Nobel, both of the Department of Psychology at Indiana University, discussed methods for building mathematical models of human memory. They spoke about the function of high-performance computing in models and testing of those models. These talks illustrated the emergence of scientific computation as a cross-disciplinary field in its own right since many of the problems and experiences presented were similar to those encountered in model-building endeavors in other disciplines.

Later that morning, Richard Golden of the School of Human Development at the University of Texas at Dallas, described a Markov random field probabilistic model in which subjects recall events in a text from memory.

James Cutting of the Department of Psychology at Cornell University used various paintings to show that the human visual system exhibits tolerance for some deviations from a Euclidean representation of space and an intolerance for others. In the pictures, perception cues were provided by occlusion, height in the visual field, relative sizes and densities, binocular disparities, motion perspectives, and the like.

Daniel Kersten of the Psychology Department at the University of Minnesota, gave a presentation of 3-D animation tools which produced realistic movies with strong perceptual cues. The goal of Kersten's work is to increase understanding of the human visual system by observing the cues and responses by viewers. Kersten applied Bayesian analysis to understand the identity of objects and their spatial relationships by cues such as cast shadows.

In the afternoon on Saturday, Mary Kaiser of the NASA Ames Research Center, Moffett Field, California, discussed how the visualization of a particlular data set-the Digital Terrain Model of Mars, which was derived from the Viking Orbiter imagery-can be optimized using our knowledge of human perception.

Jeffrey Mulligan, also from the NASA Ames Research Center, demonstrated the use of a video camera to track human eye movements. He noted that a major bottleneck is the real-time digitization and storage of large video imagery, but that recent developments in video compression hardware have made it less expensive and easier to manage these tasks. Images from the retina and the pupil can be analyzed with a basic image processing tools such as filtering, correlation, and thresholding-all of which are well-suited for implementation on vectorizing supercomputers.

Sam Williamson, Department of Physics and Center for Neural Science, New York University, presented magnetic source images (MRI) of human brain functions. Williamson is using large arrays of superconducting magnetic field sensors to map the topography of the magnetic field pattern across the human scalp.

James Anderson, Department of Cognitive and Linguistic Sciences, Brown University, focused on neural network models. He noted that although there are some good computational theories for the behavior of single neurons and that some large-scale aspects of their behavior seem lawful, there is no theory for connecting the behavior of a single neuron to the behavior of 10^11 neurons at work in the human brain. Anderson commented on the fact that as currently formulated, neural networks seem to lack essential mechanisms necessary for flexible control of the computation and neglect structure at intermediate scales of organization.

The talks began on Sunday with a presentation by Albert F. Anderson, Population Studies Center, University of Michigan, Ann Arbor. Anderson provided a real-time demonstration of a software interface for accessing and analyzing the census data via SP2 and workstation clusters. He also presented a fast data mining system for a broad spectrum of users.

Later Sunday morning, James Ramsey, Department of Psychology, McGill University, discussed a suite of techniques that promise to advance the power and scope of function estimation, while presenting interesting computational challenges.

The symposium was concluded by Patrick Suppes of Stanford University. Suppes compared the computational loads involved in the human vision system to those of analyzing turbulence in physics. He also pointed out that while human vision in naturally rich contexts is now simulated and analyzed, the work on human memory is still limited to an artificial laboratory context. There is a great need for the research on memory to play "catch-up"to research on human vision. Another area awaiting further research is that of brain magnetic resonance imaging (MRI), which involves the measurement and visualization of electromagnetic activities in the brain. He pointed out that there is not yet a meaningful correlation made between the underlying psychological processes and specific electromagnetic activities in the brain.

The symposium was opened with an overview of high-performance computer architectures presented by representatives from Cray Research Inc., Silicon Graphics Inc., and IBM. The sessions were chaired by Stephen Link, Federation of Behavioral, Cognitive, and Psychological Sciences; Al Yonas, Institute of Child Development, University of Minnesota; Cynthia Null, NASA Ames Research Center; Jay Samuels, Department of Educational Psychology, University of Minnesota; Mark Davison, Department of Educational Psychology, University of Minnesota; and Lynne Edwards Department of Educational Psychology, University of Minnesota and the Supercomputer Institute. This conference was successful in fostering interactions among speakers and the attendees on high-performance computing in the behavioral sciences. An active discussion followed each presentation. The symposium ended with closing remarks by Lynne Edwards, chair of the organizing committee, and the announcement of the symposium proceedings in early 1997.

In This Issue:

US-Japan Symposium

Behavioral Scis. Symp.

Upcoming ICPC

Undergraduate Interns

Seminar Synopses

Research Reports