Collective Knowledge: an open research SDK to share, evaluate and reuse artifacts

[mozfest-bot]

[ ID ] 515641af-a4a9-4d05-a0b5-ee6fa1a1d667

[ Submitter's Name ] Grigori Fursin [ Submitter's Affiliated Organisation ] cTuning foundation (non-profit research organization) [ Submitter's Twitter ] grigori_fursin

[ Space ] science

[ Format ] learning-lab

Description

Sharing research artifacts along with publications or at the end of research projects is gradually becoming a must for open science.

We have developed a supporting open-source technology (Collective Knowledge or CK for short) to share artifacts as reusable Python components with JSON API via GitHub; quickly prototype experimental workflows from such components as LEGO™ (such as multi-objective autotuning); automate, crowdsource and reproduce experiments; expose them to predictive analytics (scikit-learn, R, DNN), and enable interactive articles.

At this session, we will give researchers a high-level overview of CK capabilities including how to organize, share and reuse their artifacts and experimental workflows. We will also demonstrate various real CK-based experimental scenarios (such as collaboratively benchmarking and tuning Deep Neural Network algorithms and crowdsourcing compiler optimizations), already used by leading universities and research laboratories, Fortune50 companies, and computer systems' conferences (for artifact evaluation).

More information is available at:

• http://cknowledge.org/repo
• http://cTuning.org/ae

  Agenda

We plan to have a 20 minute presentation introducing Collective Knowledge concepts, 30 minutes tutorial (how to install CK, organize local code and data, share them via GitHub, create simple workflow, crowdsource experiments and apply predictive analytics), and 10 minutes for questions and answers.

Participants

We already had several CK tutorials in the past. If we have just a few participants, we will just have a more personalized approach and discussions.

Outcome

There is a growing interest towards our Collective Knowledge approach in academia and industry. Therefore, we would like to a) present CK to a wider audience b) get feedback from all participants about how improve our technology and enable open science.

For example, we initiated an open publication model where papers together with all research artifacts and experimental results become public at the time of submission to be validated and discussed by the community via SlashDot or Reddit. We believe that Collective Knowledge Framework together with such open publication model can be very useful for open science movement since it lets authors disseminate their ideas and tools much faster while avoiding unfair reviews and plagiarism: http://adapt-workshop.org/motivation2016.html

[gfursin]

If you have any questions, please do not hesitate to ask!!!

[gfursin]

And just one more thing: we have recently connected our public repository with an Android app to crowdsource experiments in computer systems' research via mobile phones, and it helps the computer engineering community collaboratively address a very old, well-known and yet unsolved problem to find best optimizations for various applications to minimize execution time, code size, compilation time, energy, etc:

• https://play.google.com/store/apps/details?id=openscience.crowdsource.experiments
• http://cknowledge.org/repo

[psyhtest]

@arlissc

Since @gfursin is presenting at ARM TechCon this week, he has passed the baton to me to present Collective Knowledge at MozFest.