[Colloquium] TALK by Markus Püschel on Wed. April 19th

[Margery Ishmael]

DEPARTMENT OF COMPUTER SCIENCE - TALK

Date: Wednesday, April 19, 2006
Time: 2:30 p.m.
Place: Ryerson 251

-------------------------------------------

Speaker: Markus Püschel

From:  Carnegie-Mellon University

Url:  http://www.ece.cmu.edu/~pueschel.

Title:  Linear Transforms: From Theory to Automatic Implementation

Abstract:

Linear transforms, such as the discrete Fourier transform, discrete  
cosine transforms, and many others, are among the most important  
numerical kernels used in signal processing and many other disciplines.

Most transforms possess a surprising number of fast algorithms, a  
fact established by the more than hundred publications on this topic.  
However, with few exceptions, the algorithms are derived through  
ingenious manipulation of the transform coefficients—a method that  
gives no insight into existence or structure of the algorithm. In the  
first part of this talk we sketch an algebraic theory of transform  
algorithms, which solves this problem for many transforms and also  
enables the derivation of many new algorithms, which could not be  
found with previous methods. (More info: www.ece.cmu.edu/~smart)

In the second part of the talk we focus on the efficient  
implementation of transforms, which is a difficult problem on fast- 
changing and increasingly complex computing platforms. We present  
Spiral, a system that aims to overcome this problem by automatically  
generating highly optimized, platform-tuned code directly from a  
problem specification. For a given transform, Spiral generates and  
explores alternative algorithms, compiles them into programs, and  
“intelligently” searches for the best match to the computing  
platform. Optimizations for the memory hierarchy and special  
instruction sets are performed automatically at the program level and  
at the mathematical level using rule-based rewriting systems.  
Experimental results show that the code generated by Spiral competes  
with, and sometimes outperforms, the best available human tuned  
library code. (More info: www.spiral.net)

Short Bio:
Markus Püschel is an Associate Research Professor of Electrical and  
Computer Engineering at Carnegie Mellon University. He received his  
Diploma (M.Sc.) in Mathematics and his Doctorate (Ph.D.) in Computer  
Science, in 1995 and 1998, respectively, both from the University of  
Karlsruhe, Germany. From 1998-1999 he was a Postdoctoral Researcher  
at Mathematics and Computer Science, Drexel University. Since 2000 he  
has been with Carnegie Mellon University. He is an Associate Editor  
for the IEEE Transactions on Signal Processing, and was an Associate  
Editor for the IEEE Signal Processing Letters, a Guest Editor of the  
Journal of Symbolic Computation, and the Proceedings of the IEEE. His  
research interests include scientific computing, compilers, applied  
mathematics and algebra, and signal processing theory/software/hardware.

***The talk will be followed by refreshments in Ryerson 255***

-------------------------------------------------------

Host:  Robert Kirby

People in need of assistance should call 773-834-8977 in advance.

For information on future CS talks: http://www.cs.uchicago.edu/events

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20060404/734036e5/attachment.htm