<\/p>\n
<\/p>\n
Continuing innovation in science and technology is vital for our society and
\nrequires ever increasing computational resources. However, delivering such
\nresources has become intolerably complex, ad-hoc, costly and error prone due to an
\nenormous number of available design and optimization choices combined with the
\ncomplex interactions between all software and hardware components.
\nAuto-tuning, run-time adaptation and machine learning based approaches
\nhave been demonstrating good promise to address above challenges for more than
\na decade but are still far from the widespread production use due to unbearably
\nlong exploration and training times, lack of a common experimental methodology,
\nand lack of public repositories for unified data collection, analysis and mining.<\/p>\n
In this talk, I will present my long-term holistic and cooperative
\nmethodology and infrastructure for systematic characterization and optimization
\nof computer systems through unified and scalable repositories of knowledge
\nand crowdsourcing. In this approach, multi-objective program and architecture
\ntuning to balance performance, power consumption, compilation time, code size
\nand any other important metric is transparently distributed among multiple
\nusers while utilizing any available mobile, cluster or cloud computer services.
\nCollected information about program and architecture properties and behavior
\nis continuously processed using statistical and predictive modeling techniques
\nto build, keep and share only useful knowledge at multiple levels of granularity.
\nGradually increasing and systematized knowledge can be used to predict most profitable
\nprogram optimizations, run-time adaptation scenarios and architecture configurations
\ndepending on user requirements. I will also present a new version of the public,
\nopen-source infrastructure and repository (cTuning3 aka Collective Mind)
\nfor crowdsourcing auto-tuning using thousands of shared kernels, benchmarks
\nand datasets combined with online learning plugins. Finally, I will discuss
\nencountered challenges and some future collaborative research directions
\non the way towards Exascale computing.<\/p>","protected":false},"excerpt":{"rendered":"
On Wednesday 18 September 2013, 11:00-12:00 INRIA Lille room B31 (new building), Grigori Fursin (INRIA, Saclay) will give a talk on “cTuning.org: systematizing program optimization using crowdsourcing and predictive modeling”. Abstract: Continuing innovation in science and technology is vital for our society and requires ever increasing computational resources. However, delivering such resources has become … <\/p>\n