Collective Knowledge (software)

From Wikipedia, the free encyclopedia
Jump to navigation Jump to search
Collective Knowledge (CK)
Collective Knowledge and cTuning logo.png
Developer(s)cTuning foundation and dividiti
Initial release2014; 5 years ago (2014)
Stable release
1.9.7 / February 11, 2019 (2019-02-11)
Written inPython
Operating systemLinux, Mac OS X, Microsoft Windows, Android
TypeKnowledge management, Artifact Evaluation, Package management system, Scientific workflow system
LicenseBSD License 3-clause
Websitegithub.com/ctuning/ck, cknowledge.org, cknowledge.org/ai, cknowledge.org/repo

The Collective Knowledge project (or CK for short) is an open-source framework and repository to enable sustainable, collaborative and reproducible research and development.[1] CK is a small, portable, customizable and decentralized infrastructure which allows researchers:

  • share their code and data as reusable Python components with unified JSON API, JSON meta information, and distributed UID via GitHub and similar services
  • quickly prototype sustainable experimental workflows from shared components as LEGO(R) (such as multi-objective autotuning[2])
  • automate, crowdsource and reproduce experiments
  • unify predictive analytics (scikit-learn, R, DNN)
  • enable interactive articles and graphs

Notable usages[edit]

Portable package manager[edit]

CK has an integrated cross-platform package manager with Python scripts, JSON API and JSON meta-description to automatically rebuild software environment on a user machine required to run a given shared research workflow.[11]

Extensible AI API[edit]

CK provides a unified and extensible JSON API for multiple DNN frameworks including Caffe and TensorFlow while optimizing them across diverse hardware (mobile devices, HPC servers) and models.[12]

Reproducibility of experiments[edit]

CK enables reproducibility of experimental results via community involvement similar to Wikipedia and physics. Whenever a new workflow with all components is shared via GitHub, anyone can try it on a different machine, with different environment and using slightly different choices (compilers, libraries, data sets). Whenever an unexpected or wrong behavior is encountered, the community explains it, fixes components and shares them back as described in.[2]

References[edit]

  1. ^ a b Fursin, Grigori; Anton Lokhmotov; Ed Plowman (January 2016). Collective Knowledge: Towards R&D Sustainability. Proceedings of the 2016 Design, Automation & Test in Europe Conference & Exhibition (DATE). Retrieved 14 September 2016.
  2. ^ a b c Grigori Fursin, Anton Lokhmotov, Dmitry Savenko, Eben Upton. A Collective Knowledge workflow for collaborative research into multi-objective autotuning and machine learning techniques, arXiv:1801.08024, January 2018 (arXiv link, interactive report with reproducible experiments)
  3. ^ HiPEAC info (page 17) (PDF), January 2016
  4. ^ Ed Plowman; Grigori Fursin, ARM TechCon'16 presentation "Know Your Workloads: Design more efficient systems!"
  5. ^ Reproducibility of Results in the ACM Digital Library
  6. ^ Artifact Evaluation for computer systems' conferences
  7. ^ EU TETRACOM project to combine CK and CLSmith (PDF)
  8. ^ Artifact Evaluation Reproduction for "Software Prefetching for Indirect Memory Accesses", CGO 2017, using CK
  9. ^ GitHub development website for CK-powered Caffe
  10. ^ Open-source Android application to let the community participate in collaborative benchmarking and optimization of various DNN libraries and models
  11. ^ http://github.com/ctuning/ck/wiki/Portable-workflows
  12. ^ Online demo of a unified CK AI API

External links[edit]

  • Development site: [1]
  • Documentation: [2]
  • Public repository with crowdsourced experiments: [3]
  • Resources related to open science: [4]
  • International Workshop on Adaptive Self-tuning Computing System (ADAPT) uses CK to enable public reviewing of publications and artifacts via Reddit: [5]
  • Android application to crowdsource experiments (such as program optimization) using mobile devices provided by volunteers via CK framework: [6]