As user demand scales for intelligent personal assistants (IPAs) such as Apple’s Siri, Google’s Google Now, Microsoft’s Cortana, and recently Amazon’s Echo, we are approaching the computational limits of current datacenter architectures. It is an open question how future server architectures should evolve to enable this emerging class of applications, and the lack of an open-source IPA workload is an obstacle in addressing this question.
In this tutorial, we present the design of Sirius, an open end-to-end IPA web-service application that accepts queries in the form of voice and images, and responds with natural language. We use this workload to investigate the implications of four points in the design space of future accelerator-based server architectures spanning traditional CPUs, GPUs, manycore throughput co-processors, and FPGAs. To investigate future server designs for Sirius, we decompose Sirius into a suite of 7 benchmarks (Sirius Suite) comprising the computationally intensive bottlenecks of Sirius. We port Sirius Suite to a spectrum of accelerator platforms and use the performance and power trade-offs across these platforms to perform a total cost of ownership (TCO) analysis of various server design points.
The purpose of this tutorial is to introduce researchers (and users) to Sirius, an open-source speech and vision based intelligent personal assistant. This tutorial will walk attendees through the various components that make an intelligent personal assistant with hands-on setup and demonstrations to familiarize users with the application and the corresponding suite available. The tutorial consists of two parts: first an overview and setup of the IPA then a deep-dive into the algorithmic components that are extracted from the end-to-end system.
- Introduction of Sirius
- Setting-up Sirius (demo)
- Sirius-Suite: a collection of IPA algorithmic components
- Running Sirius-suite (demo)
- Applications, customizations, and features
- Closing remarks, questions and comments
Saturday, March 14, 2015
Start Time: 9:00 AM
9:00 – 9:30
Introduction of Sirius
Overview of three major components of an IPA
9:30 – 10:00
Setting-up Sirius (demo)
10:00 – 10:30
10:30 – 11:00
Sirius-Suite: a collection of IPA algorithmic components
11:00 – 11:30
11:30 – 12:00
Potential applications, customizations, and features
Questions, comments, and closing remarks
Jason Mars is currently an Assistant Professor at the University of Michigan. He received his Ph.D. of Computer Science at the University of Virginia in 2012. He has been an active researcher in the areas of computer architecture, system software, and cross-layer system design within the emerging domain of cloud computing platforms. Jason has published dozens of papers in these areas and received a number of rewards and honors for excellence in his research work.
You can find out more information about Jason Mars at: www.jasonmars.org
Lingjia Tang is an assistant professor in the CSE Department at the University of Michigan. Prior to joining the University of Michigan, she was a research faculty member at UCSD CSE Department 2012-2013. She received her Ph.D. from the University of Virginia in 2012. Her research focuses on computer architecture, compiler and runtime systems, especially such systems for large scale datacenters. She publishes in the top venues in her area and has received awards for her research.
You can find out more information about Lingjia Tang at: www.lingjia.org
Johann Hauswald is a Ph.D. student in the CSE department at the University of Michigan. His research focuses on system design for emerging workloads, low-power mobile processing, and computer vision.
Learn more about Johann Hauswald at: www.web.eecs.umich.edu/~jahausw
Michael Laurenzano is a Ph.D. student in the CSE department at the University of Michigan whose research focuses on efficient warehouse-scale architectures, compilers, and runtime systems.
Learn more about Michael Laurenzano at: www.gozano.com
Yunqi Zhang is a Ph.D. student in the CSE department at the University of Michigan. His research focuses on reducing tail latency and improving energy efficiency for warehouse-scale computers.
Learn more about Yunqi Zhang at: www.web.eecs.umich.edu/~yunqi
Yiping Kang is a Ph.D. student at the University of Michigan CSE department. His current research focuses on the design space exploration of datacenters and mobile platforms for emerging workloads. www.web.eecs.umich.edu/~ypkang