DocumentCode :
3237088
Title :
Scalable Algorithms for Adaptive Statistical Designs
Author :
Oehmke, Robert ; Hardwick, Janis ; Stout, Quentin F.
Author_Institution :
University of Michigan
fYear :
2000
fDate :
04-10 Nov. 2000
Firstpage :
6
Lastpage :
6
Abstract :
We present a scalable, high-performance solution to multidimensional recurrences that arise in adaptive statistical designs. Adaptive designs are an important class of learning algorithms for a stochastic environment, and we focus on the problem of optimally assigning patients to treatments in clinical trials. While adaptive designs have significant ethical and cost advantages, they are rarely utilized because of the complexity of optimizing and analyzing them. Computational challenges include massive memory requirements, few calculations per memory access, and multiply-nested loops with dynamic indices. We analyze the effects of various parallelization options, and while standard approaches do not work well, with effort an efficient, highly scalable program can be developed. This allows us to solve problems thousands of times more complex than those solved previously, which helps make adaptive designs practical. Further, our work applies to many other problems involving neighbor recurrences, such as generalized string matching.
Keywords :
bandit models; computational learning theory; dynamic domain decomposition; dynamic programming; experimental algorithms; load balancing; memory-intensive computing; message-passing; performance analysis; sequential analysis; Algorithm design and analysis; Clinical trials; Cost function; Design optimization; Dynamic programming; Medical treatment; Multidimensional systems; Performance analysis; Standards development; Stochastic processes; bandit models; computational learning theory; dynamic domain decomposition; dynamic programming; experimental algorithms; load balancing; memory-intensive computing; message-passing; performance analysis; sequential analysis;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Supercomputing, ACM/IEEE 2000 Conference
ISSN :
1063-9535
Print_ISBN :
0-7803-9802-5
Type :
conf
DOI :
10.1109/SC.2000.10026
Filename :
1592719
Link To Document :
بازگشت