Enabling Efficient Computer Architectural and System Support for Next-Generation Deep Learning Applications

点击[ ] 时间[ 2018年12月10日 15:57] 发布人[]
报告题目:  Enabling Efficient Computer Architectural and System Support for Next-Generation Deep Learning Applications
In recent years, the artificial intelligence (AI) techniques, represented by deep neural networks (DNN), have emerged as indispensable tools in many fields. Traditionally, due to its huge compute power and scalability, the cloud data center is often the best option for training and evaluating AI applications. With the increasing computing power and energy efficiency of mobile devices, there is a growing interest in performing AI applications on mobile platforms. As a result, we believe the next-generation AI applications are pervasive across all platforms, ranging from central cloud data center to edge-side wearable and mobile devices.
However, we observe several gaps that challenge the pervasive AI applications. First, the large size of such newly developed AI networks poses both throughput and energy challenges to the underlying processing hardware, which hinders ubiquitous deployment for many promising AI applications. Second, the traditional statically trained AI model in cloud data center could not efficiently handle the dynamic data in the real in-situ environments, which leads to low inference accuracy. Lastly, the training of AI models still involves extensive human efforts to collect and label the large-scale dataset, which becomes impractical in big data era where raw data is largely un-labeled and uncategorized.
In this talk, I will present architecture and system support which enables next generation AI applications to become high efficient and intelligent. I will first introduce Pervasive AI, a user satisfaction-aware deep learning inference framework, to provide the best user satisfaction when migrating AI-based applications from Cloud to all kinds of platforms. Next, I will describe In-situ AI, a novel-computing paradigm tailored to in-situ AI applications. Furthermore, to tackle the big data challenge and achieve real intelligent (support autonomous learning), I will introduce Unsupervised AI, an unsupervised GAN-based deep learning accelerator.
Dr. Tao Li is a full professor (with preeminence professorship) in the Department of Electrical and Computer Engineering at the University of Florida. He received a Ph.D. in Computer Engineering from the University of Texas at Austin. His research interests include computer architecture, microprocessor/memory/storage system design, virtualization technologies, energy-efficient/sustainable/ dependable data center, cloud/big data computing platforms, the impacts of emerging technologies/applications on computing, and evaluation of computer systems. Dr. Tao Li received 2009 National Science Foundation Faculty Early CAREER Award, 2008, 2007, 2006 IBM Faculty Awards, 2008 Microsoft Research Safe and Scalable Multi-core Computing Award and 2006 Microsoft Research Trustworthy Computing Curriculum Award. Dr. Tao Li co-authored two papers that won the Best Paper Awards in ICCD 2016, HPCA 2011 and seven papers that were nominated for the Best Paper Awards in HPCA 2018, HPCA 2017, ICPP 2015, CGO 2014, DSN 2011, MICRO 2008 and MASCOTS 2006. Dr. Tao Li is one of the College of Engineering winners, University of Florida Doctor Dissertation Advisor/Mentoring Award for 2013-2014 and 2011-2012.
Dr. Tao Li served as a CISE program director in the National Science Foundation (NSF) during 2015-2017, directing the national research agenda in computer & system architecture, including core programs for Software and Hardware Foundation (SHF), Exploiting Parallelism and Scalability (XPS), Scalable Parallelism
in the Extreme (SPX), CISE Research Infrastructure (CRI), Faculty Early CAREER Development (CAREER), CISE Research Initiation Initiative (CRII), and Expeditions in Computing (EIC) programs. Dr. Tao Li is an IEEE Fellow.
新闻动态更多 >>
友情链接 北京航空航天大学国内外大学上级机关驻华使领馆百度谷歌后台管理

电话:010-82338050 传真:010-82339376 地址:北京海淀区学院路37号 邮编:100191
International Division BUAA ? 2007/privacy Polilcy