Accelerating Neural Computation

已经在生物学中找到其功能的灵感或模拟实际生物系统功能的计算系统已经被引入多年，并且随着技术的进步允许部署高性能计算系统，最近已经看到它们的发展加速。在这项研究计划中，我们将专注于开发加速神经网络所需计算的方法。神经网络可以被视为（i）在分类，建模和学习中找到应用的人工神经网络，以及（ii）更接近中枢神经系统结构和功能的尖峰神经形态系统。我们的长期目标是设计一个系统，可以精确地模拟大脑的一部分在真实的或接近真实的时间。这样的系统可以用于生物学研究，以测试假设，特别是与疾病有关的假设，并更好地了解结构对功能的影响。这样的目标是新颖的，因为目前大多数神经形态系统研究都集中在开发可扩展的复杂系统上，通常使用简化的神经模型和不能准确代表生物学的互连。我们的方法在计算上具有挑战性，但其复杂性是易于处理的。另一个目标是开发提高泛化ANN性能的系统。* 预计这项研究的成功完成将导致高级研究，并有助于理解复杂的神经系统。* 在人工神经网络领域，拥有高效的计算环境将允许对复杂系统和分类器进行建模，并将有助于大数据分析。* 在神经形态系统领域，能够在真实的或接近真实的时间内准确模拟大脑结构，将允许研究人员进行实验并了解这些大脑结构的功能。这样的系统有可能成为虚拟实验室，对这样的大脑结构进行实验，刺激它们的功能，探索学习是如何发生的，并了解一些退化性疾病是如何表现出来的。

Computing systems that have found the inspiration of their function in Biology or model the functioning of actual biological systems have been introduced for many years and have seen their development accelerating recently as advances in technology have allowed the deployment of high performance computing systems.***In this research program, we will focus on developing methods of accelerating the computations needed in neural networks. Neural networks can be viewed as (i) Artificial Neural Networks that have found applications in classification, modeling and learning, and (ii) spiking neuromorphic systems that closer resemble the structure and functioning of the central nervous system.***Our long term objective is to devise a system that can accurately emulate a part of the brain in real or near-real time. Such a system can be used in biological research to test hypotheses especially as related to diseases, and to better understand the influence of the structure on function. Such a goal is novel, as most of the current neuromorphic systems research is focusing on developing complex systems that scale, often using simplified neural models and interconnects that do not accurately represent biology. Our approach is computationally challenging, but its complexity is tractable. An additional goal is the development of systems enhancing the performance of generalizing ANN. ***It is expected that the successful completion of this research will lead to advanced study, and contribute to the understanding of complex neural systems. ***In the field of artificial neural networks, having an efficient computational environment will allow the modeling complex systems and classifiers and it will contribute to big data analytics. ***In the field of neuromorphic systems, having the ability to accurately simulate brain structures in real or near-real time, will allow researchers to experiment and understand the functioning of these brain structures. Such systems have the potential of becoming virtual laboratories to experiment on such brain structures, prod their functionality, explore how learning occurs and understand how some degenerating diseases manifest themselves.