CONNECTIONIST MODELING IN THE NEUROLOGY OF AGING

点击翻译按钮获取中文摘要

The neurological diseases of aging lack reliable clinical markers in the physical and mental status examination. Differential diagnosis between normal aging and early senile dementia of the Alzheimer type (SDAT), between dementia and pseudodementia and between SDAT and multi-infarct dementia is problematic. The development of clinical criteria depends on a better understanding of brain function and of functional response to damage. Theories of higher cortical function, derived from studies of stroke, are dependent on localization of function and are thus not applicable to the diffuse diseases of aging. Connectionism is a method of neurologic modeling that uses distributed representations and parallel processing. These models are physiological in that they use several brain-like skills: learning, ability to generalize, disambiguation of inputs, retrieval of memory by content not location, and graceful degradation. The goal will be to use neural network models to adduce clinical markers for specific diseases and for prediction and testing of potential neurotransmitter therapies. These models will focus clinical trials by correlating neurotransmitter effects with clinical findings. A proposal is presented for a training program in connectionism oriented to basic research in the clinical neurology of aging. Phase I of the program will involve course work in biophysics, computer science, statistics and mathematics as well as laboratory exposure in neurophysiology, clinical work in dementia and experience with ongoing connectionist research in Dr. Sejnowski's laboratory. In phase II, connectionist methods will be used to design systems that reflect the specific diseases of aging. Clinical and neurophysiological data will be used to constrain the models. The design is expected to use diffuse associative connectivity with specific synaptic connectivity added. These models will be manipulated to simulate particular diseases.