Presidential Young Investigators Award: Fast Neural Network Like Optimization Techniques

P.I.S目前对大规模并行、干扰和集体计算、人工神经系统、学习机、联想神经存储器、光子计算和光学神经网络实现的研究感兴趣。在神经网络领域，他计划调查各种联想神经记忆架构和相关记录/学习算法的性能。他为联想记忆引入了新的高性能记录/学习算法，目前正在为各种记录/学习算法研究这种神经记忆的动力学。在计算和神经网络实验室，他和他的学生正在开发用于聚类和无监督模式分类的自我优化的无监督学习神经网络结构。在另一个项目中，神经网络结构将被用于解决二维和更高维的复杂最优路径寻找问题。他还将研究人工神经系统的光学实现。他拥有第一个集成光学阈值门(“光神经元”)的专利，并开发了光纤交叉棒网络制造技术，这些技术在光神经网络实现中具有潜在的实用价值。他最近完成了一个基于光纤的高性能动态联想神经存储器原型的设计和实现。他还对神经网络在信号处理和并行计算中的应用感兴趣。在与韦恩州立大学神经科的研究人员的共同努力下，他正在开发一种基于神经网络的技术，用于临床诊断目的的临床肌电(EMG)的自动量化。他未来的研究将集中在高性能动态人工神经网络的自适应无监督合成技术上。他还将研究如何通过自我优化的神经架构来加快多层神经网络的学习速度。他未来的研究还将涉及提出和实现大规模高性能光纤互联动态联想神经存储器和其他神经网络范例。

The P.I.'s currently interested in research in massively parallel, disturbed, and collective computations, artificial neural systems, learning machines, associative neural memories, photonic computing, and optical neural network implementations. In the area of neural networks, he plans to investigate the performance of various associative neural memory architectures and related recording/learning algorithms. He has introduced new high-performance recording/learning algorithms for associative memories, and is currently investigating the dynamics of such neural memories for various recording/learning algorithms. In the Computation and Neural Networks Laboratory, he and his students are developing unsupervised learning self-optimizing neural net architectures for clustering and unsupervised pattern classification. In another project, neural net architectures will be employed in solving complex optimal path finding problems in two- and higher-dimensions. He will also be investigating the optical implementations of artificial neural systems. He holds a patent for the first integrated optical threshold gate ("optical neuron") and has developed fiber optic cross-bar network fabrication techniques, which are potentially useful in optical neural network implementations. He has recently completed the design and implementation of a fiber optic-base high-performance dynamic associative neural memory prototype. He is also interested in the application of neural networks in signal processing and parallel computing. In a joint effort with researchers at Wayne State's Neurology Department, he is developing a neural-network-based technique for automatic quantification of the clinical Electromyogram (EMG) for clinical diagnostic purposes. His future research will focus on adaptive unsupervised synthesis techniques for high- performance dynamic artificial neural networks. He will also be investigating ways of speeding up learning in multiple-layer neural nets through self-optimizing neural architectures. His future research will also involve proposing and implementing large scale high- performance fiber optic interconnected dynamic associative neural memories and other neural network paradigms.