Parallel Simulation of Neurons

点击翻译按钮获取中文摘要

My long term goal is to develop parallel discrete models for computationally challenging, significant problems because (1)discrete event models facilitate detailed modeling and (2) can be executed in parallel. I am proposing, for this DG, the parallel discrete event simulation of the interaction between chemical and electrical models of a neuron. The project will be done in collaboration with Dr. Lytton, a neuroscientist at the Downstate Medical School (SUNY) and Dr. Hines, a neuroscientist at Yale University. NEURON, a widely used simulator in computational neuroscience will make use of this research. Benefits of this research are (1) an enhanced understanding of the neuron which can lead to treatments for neurological disorders (2) progress in development of scalable hybrid discrete event simulations. The membrane of a neuron is semi-permeable and has channels within it which control the flow of ions between the exterior and the interior of the cell body.Detailed electrical models, based upon well-known electrical laws describe many of the properties of a neuron. However, an assumption behind these models, that intracellular and extracellular ionic concentrations are constant is actually violated by calcium ions, which play a significant role in intracellular signaling. Hence it is necessary to model ionic concentrations in order to correctly model the electrical response of a neuron. Key computational challenges are the fine-grained nature of event processing and the need to maintain a very large state history. I expect to leverage the extensive research done in my group on dynamic load balancing algorithms and on the application of AI methods to control optimistic simulation (computation of window size, dynamic load balancing). The subsequent integration of the electrical (continuous) and chemical models (discrete event) will follow.We will continue our work on the n-body problem for stellar systems, making use of optimistic simulation along with collision detection algorithms to investigate the problem of bodies (rubble piles) which result from collisions and which are held together by gravity. Benefits include a better understanding of planetary formation and the development of scalable algorithms for the n-body problem.