Deep Neural Networks (DNNs) have become the computationaltool of choice for many applications relevant to mobile devices.However, given their high memory and computational demands, running themon mobile devices has required expert optimization or custom hardware.We present a framework that, given an arbitrary DNN, compiles it down toa resource-efficient variant at modest loss in accuracy. Further, weintroduce novel techniques to specialize DNNs to contexts and to shareresources across multiple simultaneously executing DNNs. Finally, wepresent a run-time system for managing the optimized models we generateand scheduling them across mobile devices and the cloud. Using the challenging continuous mobile vision domain as a case study, we showthat our techniques yield very significant reductions in DNN resourceusage and perform effectively over a broad range of operatingconditions.

Seungyeop Han is a Ph.D candidate in the Department of Computer Science and Engineering at University of Washington. His research interests are in the broad area of distributed systems and computer networks, and also includerelated topics like security and privacy. He has published papers in the premier conferences such as SIGCOMM, Ubicomp, NSDI, CCS, USENIX Security, NIPS,ATC, CHI and WWW.He received KFAS fellowship in computer science (2010-). Prior to studying in UW, he worked for Naver as a software engineer for 3 years. He received his B.S. (2005) and M.S. (2007) in Computer Science from KAIST.