Collective Mind Node

tối ưu hóa chương trình hợp tác và học máy sử dụng thiết bị di động

THIS APPLICATION IS NOW DEPRECATED! Download brand new application for universal experiment crowdsourcing (including GCC and LLVM crowd-tuning) here:

* https://play.google.com/store/apps/details?id=openscience.crowdsource.experiments

The new version is based on open Collective Knowledge research SDK:

* http://github.com/ctuning/ck

You can see all public optimization results here:

* http://cTuning.org/crowd-results

=======================================================================

Tuning applications for rapidly evolving hardware or improving optimization heuristic of a new compiler is becoming complex, ad-hoc, costly and error prone due to an enormous number of available optimization choices combined with the complex interactions between all software and hardware components.

Since 2008, we are developing a methodology and supporting technology to crowdsource compiler auto-tuning (iterative compilation) across mobile devices and make it practical using our open-source cTuning technology (http://cTuning.org). In our approach, multi-objective codelet, program and architecture tuning to balance performance, power consumption, compilation time, code size and any other important metric is continuously distributed among multiple participants.

We hope that our approach will help to automatically improve default optimization heuristics of latest compilers including GCC, LLVM, Open64 and any other, or predict optimizations for new programs using collected knowledge and machine learning to maximize utilization of modern computer devices. Interestingly, we also hope that it will help to improve academic research and switch focus from publishing numerous non-reproducible papers to sharing research material and experimental results that can be collaboratively validated and improved by the community.

This application, supported by non-profit cTuning foundation, is currently a research proof-of-concept to support and validate our idea. It helps us collect various performance statistics from participants to validate machine learning techniques. This is an on-going project, so please be patient! Feedback and suggestions are welcome!

More info:

* Vision publications: http://arxiv.org/abs/1506.06256 , https://hal.inria.fr/hal-01054763

* Community website: http://cTuning.org

* Live Repositories: http://c-mind.org/repo

* Our related artifact evaluation initiative (to start collecting more public benchmarks and data sets): http://cTuning.org/ae

* Author page: http://fursin.net

We are currently raising further funding to continue this project and related public research activities. If you are interested, do not hesitate to contact Grigori Fursin ([email protected]) for more details!