To visit Dr. Maida's personal web page, click
here
|
Associate Professor, Computer Science.
B.A., Mathematics, State University of New York at Buffalo, 1973; Ph.D.,
Psychology, State University of New York at Buffalo, 1980; M.S., Computer
Science, State University of New York at Buffalo, 1981; Postdoctoral
Fellowship, Cognitive Science, Brown University; Postdoctoral Fellowship,
Cognitive Science, University of California, Berkeley.
|
|
| Dr. Maida's research interests are in artificial intelligence and cognitive science. His current and primary research effort develops multimedia tools to interactively visualize neuronal simulations of small systems of neurons such as the rat hippocampus. The hippocampus is an interesting part of the brain because, in humans, it is needed to add new information to long-term memory. In rats, the hippocampus maintains representations that allow the rat to remember how to return to previously visited food locations. More details are known about neuronal structure and function in the rat than in humans because neuroscientists can take more direct measurements (e.g., inserting electrodes into the brain to measure neuronal activity). Thus, more information is available to tell us how to "wire up" virtual neurons to build a simulation. Neuroscientists have not yet produced a complete wiring diagram of the hippocampus. Consequently, one must use many forms of indirect evidence to first piece together a plausible model of hippocampal function and then build a simulation to generate predictions. | The predictions help one to judge whether the model is correct and may also help neuroscientists design better experiments. Complications arise when building such simulations. First, one does not have the luxury of using existing mathematical models found in the artificial neural network literature. Though such models are historically "brain inspired," they do not match the structural details of the brain. Alternatively, if one builds a simulation directly from the incomplete wiring diagram of the known nervous system, the simulation's internal causal structure will remain a mystery. The simulation will likely generate odd and counter-intuitive predictions that pose the awkward problem of determining whether the implementation has a bug or whether the implementation is correct but the model really does make the prediction. These problems motivated his research to develop simulation tools to allow cognitive and neuroscientists to study hippocampal models using exploratory visualizations. The technology will allow a user to begin with an idealized mathematical model | and develop intuitions about causes underlying its function. The user may then incrementally add biological realism to the model while comparing visualizations over the series of refinements in order to judge effects of modifying the model. His earlier research studies description-based communication protocols for autonomous agents. The agents are assumed to cooperatively explore an unknown environment such as the Martian surface. They independently discover or rediscover new objects (such as geologically significant rocks) and must exchange information about these discoveries. The description-based approach derives from analogy to a similar communication problem in human language where humans use descriptions to refer to objects that do not have names (such as that plastic thing on the tip of your shoe lace). |