Projects per year
Abstract
Estimation of Distribution Algorithms
(EDAs) are a popular approach to learn
a probability distribution over the \good"
solutions to a combinatorial optimization
problem. Here we consider the case where
there is a collection of such optimization
problems with learned distributions, and
where each problem can be characterized
by some vector of features. Now we can
dene a machine learning problem to predict
the distribution of good solutions q(sjx) for
a new problem with features x, where s
denotes a solution. This predictive distribution
is then used to focus the search. We
demonstrate the utility of our method on a
compiler optimization task where the goal is
to nd a sequence of code transformations to
make the code run fastest. Results on a set
of 12 dierent benchmarks on two distinct
architectures show that our approach consistently
leads to signicant improvements in
performance.
| Original language | English |
|---|---|
| Title of host publication | ICML '06 Proceedings of the 23rd international conference on Machine learning |
| Publisher | ACM |
| Pages | 121-128 |
| Number of pages | 8 |
| ISBN (Print) | 1-59593-383-2 |
| DOIs | |
| Publication status | Published - 2006 |
Fingerprint
Dive into the research topics of 'Predictive Search Distributions'. Together they form a unique fingerprint.Projects
- 2 Finished
-
Compilers that Learn to Optimise (COLO)
O'Boyle, M. (Principal Investigator) & Williams, C. (Co-investigator)
16/07/04 → 15/09/07
Project: Research
-
PASCAL: Pattern analysis, statistical modelling and computational learning
Williams, C. (Principal Investigator)
1/12/03 → 30/09/08
Project: Research
Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver