CSE 590G -- the Architecture Lunch (Fall 2001)

General purpose architectures provide instructions sets and hardware that can execute any application with decent performance. At the other end of the spectrum sit ASICs, whose hardware is designed to fit a particular application. ASICs achieve better performance than general purpose processors on their particular application, but often lack the flexibility to allow changes to their application or execute other, even related, applications.

Last spring in 590G (the architecture seminar, aka the architecture lunch) we read papers on domain-specific architectures, that sit somewhere between general purpose architectures and ASICs, and that provide a performance or power boost to emerging application domains by extending the instruction set or portions of the microarchitecture.

This autumn quarter we would like to get a better understanding of the application domains themselves, with a view toward developing a research project on architectural support for interesting domains and/or building a domain-specific processor. We have identified several domains for which domain-specific architectures have not yet been developed or are not what we would call mature and for which there are strong research groups on campus with whom we could collaborate.

The domains may include:

  • the "natural I/O" areas of speech recognition (Jeff Bilmes in EE) and vision (Steve Seitz and Brian Curless),
  • AI algorithms, particularly machine learning (Pedro Domingos, Chris Diorio and students) and robotics (Dieter Fox and students),
  • network processing, e.g., software radios (Tom Anderson, David Wetherall and students),
  • computational biology (Larry Ruzzo, Martin Tompa, Rimli Sengupta and students),
  • genomics (Leroy Hood's Institute for Systems Biology and the group of university departments that comprise Computation Molecular Biology),
  • ubiquitous computing in the biology lab (Larry Arnstein and the Cell Systems Initiative in the medical school).
  • cell phones
  • what else? 

    •  

     

    Our plan is to tackle a domain each week by reading a paper and discussing the domain's implications for architectural design with local faculty and students working in the domain or an outside speaker. The 590G web page: http://www.cs.washington.edu/education/courses/590g/ will contain a schedule for the discussions.

    We encourage both architecture students and students interested in the particular domains to attend. 590G meets on Tuesdays at 12:30 in EE1 037.

    Bring your lunch!

    Jean-Loup Baer, Carl Ebeling, Susan Eggers, Scott Hauck, Mark Oskin

    The Weekly Schedule

    Below is the discussion schedule. Entries in bold are for sure; the entry in red is the next discussion.
     
     
    Week
    Domain
    Special Participants & Contact 
    1 
    October 2 
    		 Discussion of which domains we want to tackle and when  Everyone! 
    2 
    October 9 
    		 Sherwood and Calder, ISCA-2001  Steve Swanson
    3 
    October 16 
    		 speech recognition  Jeff Bilmes from EE will give a talk on speech recognition algorithms and lead the discussion afterwards. Jeff's slides (pdf) 
    computational biology  Martin Tompa will speak in 519 on "Identifying Regulatory Motifs in DNA Sequences", 3:30 in 134. 
    4 
    October 23 
    		 vision: 3D modeling from images, and a bit of Lucas/Canade motion estimation and/or high dynamic range imaging  Steve Seitz. Steve's slides (ps) 
    5 
    October 30 
    		 networking 
    Vanu's recommended paper (ps) & a somewhat outdated overview on software radios (ps)     		 "Software Radios", Vanu Bose from Vanu Inc., 134 Sieg at 3:30. Vanu is speaking in the department colloquium. 
    6 
    November 6 
    		 machine learning algorithms  (tutorial paper)  Pedro Domingos 
    7 
    November 13 
    		 FPGA implementations of machine learning algorithms  Miguel Figueroa 
    8 
    November 20 
    		 Gene sequence analysis  Tim Hunkapiler, Discovery Biosciences
    9 
    November 27 
    		 XML   Dan Suciu
    10 
    December 4 
    		 Discussion of which domains seem the most promising for further research.  Everyone! 

    Other possibilities:

    Last modified on Friday, 26-Oct-2001 16:21:05 PDT