Paper Info
Title
Predictive search distributions
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 distri- bution 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 dieren t benchmarks on two distinct architectures show that our approach consis- tently leads to signican t improvements in performance.
Year
DOI
Venue
2006
10.1145/1143844.1143860
ICML
Keywords
Field
DocType
combinatorial optimization problem,popular approach,compiler optimization task,optimization problem,predictive search distribution,new problem,distribution algorithms,good solutions q,code transformation,probability distribution,predictive distribution,estimation of distribution algorithm,machine learning,compiler optimization
Estimation of distribution algorithm,Computer science,Vector optimization,Quadratic assignment problem,Combinatorial optimization,Cross-entropy method,Theoretical computer science,Artificial intelligence,Random optimization,Optimization problem,Machine learning,Metaheuristic
Conference
ISBN
Citations 
PageRank 
1-59593-383-2
3
0.62
References 
Authors
6
6
Name
Order
Citations
PageRank
Edwin V. Bonilla1100853.32
Christopher K. I. Williams26807631.16
Felix V. Agakov344234.22
John Cavazos458426.93
John Thomson550523.11
Michael F. P. O'Boyle6110165.55