Check the preview of 2nd version of this platform being developed by the open MLCommons taskforce on automation and reproducibility as a free, open-source and technology-agnostic on-prem platform.
Summer CK digest (blog article, 2016-09-15)
Grigori Fursin 

[ Optimization repository ] [ GCC ] [ LLVM ] [ MILEPOST optimization predictor ]
[ Andrew Ng's ML course ] [ Google ML crash course ] [ Facebook Field Guide to ML ]

Dear colleagues,

Hope you had a nice and relaxing summer!

There have been many news related to our Collective Knowledge framework this summer, so I would like to share some of them with you.


We would like to thank Microsoft for providing a 1-year grant to host repository at Microsoft Azure cloud!

We will present a CK-based project with ARM at ARM TechCon'16 in Santa Clara in October (see schedule, DATE'16 paper and CPC'15 paper). We will demonstrate Workload Knowledge, an open framework for gathering and sharing knowledge about system design and optimization using real-world workloads. Powered by 3 open-source projects (ARM's Workload Automation, cTuning's Collective Knowledge and Jupyter Notebooks), Workload Knowledge will dramatically accelerate innovation in computer engineering and lead to designing highly efficient systems.

We will announce various Collective Knowledge awards for the top contributors sharing workloads, data sets, tools, autotuning plugins, predictive models and optimization knowledge at ARM TechCon. Active student contributors will have a priority for internships at dividiti! Hurry up to try CK and join our growing community!

Program crowd-tuning

We would like to thank all participants for testing our CK-powered GCC/LLVM crowd-tuning approach using spare mobile phones, tablets, laptops and cloud servers. We now have more that 300 distinct Android, Windows, Linux and MacOS-based platforms participated in experiment crowdsourcing (nearly 150 distinct CPU, and ~ 50 distinct GPUs). You can find all meta shared at GitHub.

Furthermore, the community collected and shared numerous distinct GCC and LLVM optimizations for more than 70 (CPU,compiler version) tuples across ~140 shared workloads. It opens up many interested opportunities for practical research in machine-learning based autotuning, run-time adaptation and SW/HW co-design!

CK improvements

Above collaborative optimization helped us stabilize CK framework including Android mobile app. Thanks to the community contributions and fixes, CK autotuning now supports MacOS.

We also added support for continuous integration frameworks for Linux and Windows, for Jupyter notebooks, and for Docker!

We have released new Collective Knowledge Framework V1.8.1 (available via pip):


We have also released Android application V2.2 with sources to help you participate in various experiment crowdsourcing using Android-based mobile devices and IoT devices:


Finally, we provided a new documentation with various Getting Started Guides:


Open Science and Reproducible Research

* Congratulations to Dr. Abdul Memon (my last PhD student) for successfully defending his thesis "Crowdtuning: Towards Practical and Reproducible Auto-tuning via Crowdsourcing and Predictive Analytics" in the University of Paris-Saclay. Most of the software, data sets and experiments are not only reproducible but also shared as reusable and extensible components via Collective Mind and CK!

* We have moved our "Open Science" wiki with related resource to CK GitHub.

* We have helped with Artifact Evaluation for PACT'16!

* We continue discussions with ACM colleagues about how to enable collaborative and reproducible R&D across all SIGs.

* Please, check this interview with Dr. Anton Lokhmotov (CEO of dividiti) above how CK can help enable efficient, reliable and cheap computing everywhere!


After nearly 20 years, we are finally moving back to AI research and have several projects related to unifying benchmarking, tuning and access to DNN networks via CK:


We also work on a unification of benchmarking and multi-objective autotuning across remote devices; adding more workloads in the CK format; adding more machine-learning based autotuning strategies; crowdsourcing compiler bug detection; crowdtuning OpenCL and CUDA BLAS libraries; improving documentation further; unifying predictive analytics including access to DNN frameworks via CK web service.

See our assorted plans here. You can also check open tickets at GitHub pages of and

The community

Since there is a growing number of CK-powered collaborative and reproducible projects as well as CK users and participants in experiment crowdsourcing, we strongly suggest you to participate in public discussions via this public mailing list or LinkedIn group. This will help the community share knowledge and experience about CK while avoiding common pitfalls!

Have a very productive Fall and looking forward discussing CK projects with you,
Grigori and the Collective Knowledge team


Please log in to add your comments!
If you notice any inapropriate content that should not be here, please report us as soon as possible and we will try to remove it within 48 hours!