Bandit-Based Optimization on Graphs with Application to Library Performance Tuning

Abstract : The problem of choosing fast implementations for a class of recursive algorithms such as the fast Fourier transforms can be formulated as an optimization problem over the language generated by a suitably defined grammar. We propose a novel algorithm that solves this problem by reducing it to maximizing an objective function over the sinks of a directed acyclic graph. This algorithm valuates nodes using Monte-Carlo and grows a subgraph in the most promising directions by considering local maximum k-armed bandits. When used inside an adaptive linear transform library, it cuts down the search time by an order of magnitude compared to the existing algorithm. In some cases, the performance of the implementations found is also increased by up to 10% which is of considerable practical importance since it consequently improves the performance of all applications using the library.
Document type :
Conference papers
ICML, Jun 2009, Montréal, Canada. 2009
Liste complète des métadonnées


https://hal.inria.fr/inria-00379523
Contributor : Arpad Rimmel <>
Submitted on : Tuesday, April 28, 2009 - 8:20:37 PM
Last modification on : Thursday, February 9, 2017 - 3:55:00 PM
Document(s) archivé(s) le : Thursday, June 10, 2010 - 7:44:00 PM

File

icmlspiral.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : inria-00379523, version 1

Collections

Citation

Frédéric De Mesmay, Arpad Rimmel, Yevgen Voronenko, Markus Püschel. Bandit-Based Optimization on Graphs with Application to Library Performance Tuning. ICML, Jun 2009, Montréal, Canada. 2009. <inria-00379523>

Share

Metrics

Record views

120

Document downloads

125