|
||
With funding from the NSF's Human Cognition and Perception Program, part of the BCS (Behavioral, & Cognitive Sciences) division, Dr. Anderson, Chair of Brown University's Department of Cognitive and Linguistic Sciences and colleagues have recently studied neural network modelling of human reaction times. For more than a century, psychologists have studied the patterns seen in the time it takes a person to produce an answer to a problem in an effort to better understand the details of mental operation. This venerable technique benefits from new approaches using Dr. Anderson's nonlinear BSB neural network combined with computer simlulations. A major part of this project involves looking at the way humans solve simple arithmetic problems, a surprisingly difficult task for both humans and artificial neural networks. Instead of actually computing the answers to problems the way a digital computer would, humans (and networks) seem to learn elementary arithmetic very differently, by memorizing facts and estimating answers. Although these human strategies can be error-prone and slow, natural extensions of them can give rise to the poweful but poorly understood faculty of mathematical intuition as well as the ability to reason "intuitively" about complex systems. Another NSF funded project in which Dr. Anderson is involved is funded through the LIS (Learning and Intelligent Systems) initative which promotes studies on intelligence in humans, animals and artifical systems. Anderson and colleagues are involved in a study on "Adaptive Cortical Computation in the Visual Domain" which is investigating the long range spatial interactions among neurons during visual processing. The neural network model used in this study is called informally the "network of networks" and is an attempt to bridge the huge gap in scale between single processing elements (neurons) and entire brain regions that may contain hundreds of millions of cooperating neurons. This model suggests that neurons can work together to build larger and larger functional groupings and that larger, and even larger groupings may form from the smaller ones using similar rules of formation. In his years of NSF-funded research James Anderson has proven repeatedly that, in his own words, "cognitive science can, in fact, be immensely practical in the right situation." Network models similar to those he and his colleagues first developed with NSF support to simulate the human nervous system have become the foundation for the artificial neural networks now used routinely in many pattern based applications such as credit verification, medical diagnosis, and speech recognition. Anderson has also collaborated with companies such as Texas Instruments to improve military electronics. One project required a means of analyzing a confusing flood of radar signal data (see sidebar). The radar data was processed by a neural network designed to simplify the complex, as humans do, by breaking information into manageable blocks of data. Humans use "concepts" as a way of simplifying and understanding a complex environment. The techniques used for radar analysis were based directly on neural network models for human concept formation. He and colleagues at Distributed Data Systems, Inc. have used this idea to develop a "smart" radar analysis system for the U.S. Navy. The NSF has also supported Anderson and colleagues by providing funds for the purchase of powerful modern computer facilities for these neural network simulations as well as for related work helping to understand the mechanics of the human mind. For more information please see:
|
|
|
Dr. Anderson's Web page at: http://biomedcs.biomed.brown.edu/NeuroBrochure/Faculty/Anderson.html This research is supported by the Human Cognition and Perception Program and the Learning and Intelligent Systems Initative. |
All photos and illustrations are copyright© of their respective owners and may not be used without permission. |
|