The prototyping phase of the CK technology is over!
2 views
Skip to first unread message
Grigori Fursin
Jun 26, 2020, 5:26:23 AM6/26/20
to Collective Knowledge, ctuning-discussions, Artifact Evaluation for systems and ML conferences
Dear all,
I hope you are fine during these hard times!
I am very excited to announce that after several years of hard work we have completed the prototyping phase of the Collective Knowledge technology (CK): https://arxiv.org/abs/2006.07161 .
My goal was to make it easier to co-design efficient software and hardware for emerging workloads, reproduce AI, ML & systems research, and deploy novel techniques in production with the help of portable workflows, reusable artifacts, and DevOps principles.
As a proof-of-concept, we have demonstrated the possibility to assemble portable and reproducible workflows that can automatically adapt to continuously changing software, hardware, models, and data sets. We also showed that it was possible to reuse research code, data, and best practices in the form of open CK APIs, JSON meta descriptions, and DevOps principles instead of reinventing the wheel.
We even enabled "live" research papers where experiments were crowdsourced with the help of volunteers and results were validated by the community. Such an approach helped to connect researchers and practitioners to collaboratively benchmark, optimize and co-design ML models, software and hardware while automatically trading off speed, accuracy, energy, size, and different costs using our open repository of knowledge with live scoreboards: https://cKnowledge.io/reproduced-results .
However, CK is still a prototype and there is a lot to be improved particularly in terms of simplicity, usability, and standardization. That is why I decided to go offline this summer to reflect on all the feedback I have received during the past few years, brainstorm the new version, and prepare new projects.
Please stay tuned and feel free to check these CK resources in the meantime:
* Project website: https://cKnowledge.org
* Use-cases: https://cKnowledge.org/partners
* Portal with reusable components: https://cKnowledge.io/browse
* Crowd-benchmarking demo: https://cKnowledge.io/test
* GitHub: https://github.com/ctuning/ck
Take care,
Grigori
I hope you are fine during these hard times!
I am very excited to announce that after several years of hard work we have completed the prototyping phase of the Collective Knowledge technology (CK): https://arxiv.org/abs/2006.07161 .
My goal was to make it easier to co-design efficient software and hardware for emerging workloads, reproduce AI, ML & systems research, and deploy novel techniques in production with the help of portable workflows, reusable artifacts, and DevOps principles.
As a proof-of-concept, we have demonstrated the possibility to assemble portable and reproducible workflows that can automatically adapt to continuously changing software, hardware, models, and data sets. We also showed that it was possible to reuse research code, data, and best practices in the form of open CK APIs, JSON meta descriptions, and DevOps principles instead of reinventing the wheel.
We even enabled "live" research papers where experiments were crowdsourced with the help of volunteers and results were validated by the community. Such an approach helped to connect researchers and practitioners to collaboratively benchmark, optimize and co-design ML models, software and hardware while automatically trading off speed, accuracy, energy, size, and different costs using our open repository of knowledge with live scoreboards: https://cKnowledge.io/reproduced-results .
However, CK is still a prototype and there is a lot to be improved particularly in terms of simplicity, usability, and standardization. That is why I decided to go offline this summer to reflect on all the feedback I have received during the past few years, brainstorm the new version, and prepare new projects.
Please stay tuned and feel free to check these CK resources in the meantime:
* Project website: https://cKnowledge.org
* Use-cases: https://cKnowledge.org/partners
* Portal with reusable components: https://cKnowledge.io/browse
* Crowd-benchmarking demo: https://cKnowledge.io/test
* GitHub: https://github.com/ctuning/ck
Take care,
Grigori
Davide Del Vento
Jun 29, 2020, 7:05:43 PM6/29/20
to Grigori Fursin, ctuning-discussions, Artifact Evaluation for systems and ML conferences, collective...@googlegroups.com
Hi Grigori,
Congratulations on reaching this milestone and taking this decision to move forward. I've been really impressed with how your project evolved, with the bold steps you have taken along the way, and with what you have achieved. I am happy to have contributed in that minuscule way with our conversations at my conference, and I am sorry for not having had the opportunity to be more involved...
That being said, I am sure the future way will be exciting, and I expect you will forge a great path forward.
Best,
Davide Del Vento,
NCAR Computational & Information Services Laboratory
Consulting Services Software Engineer
http://www2.cisl.ucar.edu/uss/csg/
SEA Chair http://sea.ucar.edu/
office: Mesa Lab, Room 55G
phone: (303) 497-1233
NCAR Computational & Information Services Laboratory
Consulting Services Software Engineer
http://www2.cisl.ucar.edu/uss/csg/
SEA Chair http://sea.ucar.edu/
office: Mesa Lab, Room 55G
phone: (303) 497-1233
--
You received this message because you are subscribed to the Google Groups "Collective Knowledge Technology" group.
To unsubscribe from this group and stop receiving emails from it, send an email to collective-knowl...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/collective-knowledge/a69fb1ce-ca33-900e-9ffd-26bceede1ccc%40cTuning.org.
Grigori Fursin
Jul 6, 2020, 7:31:22 AM7/6/20
to Davide Del Vento, ctuning-discussions, Artifact Evaluation for systems and ML conferences, collective...@googlegroups.com
Thank you very much, Davide, for your nice words!
And thank you so much for the great discussions we had for nearly 10 years (!) about how to automate the design of efficient software and hardware for scientific workloads! I would also like to stress the importance of the SEA conference that you were organizing for so many years - it helps to connect scientists and engineers to discuss the best practices for software engineering and how to enable collaborative and reproducible R&D. These discussions helped me when designing the Collective Knowledge framework particularly since I had a very tough mission to make CK useful for both researchers and practitioners with very different and sometimes contradictory requirements ;) ! Of course, there is still a lot to be improved and I hope to find resources to continue these developments later this year ...
Take care and keep in touch,
Grigori
And thank you so much for the great discussions we had for nearly 10 years (!) about how to automate the design of efficient software and hardware for scientific workloads! I would also like to stress the importance of the SEA conference that you were organizing for so many years - it helps to connect scientists and engineers to discuss the best practices for software engineering and how to enable collaborative and reproducible R&D. These discussions helped me when designing the Collective Knowledge framework particularly since I had a very tough mission to make CK useful for both researchers and practitioners with very different and sometimes contradictory requirements ;) ! Of course, there is still a lot to be improved and I hope to find resources to continue these developments later this year ...
Take care and keep in touch,
Grigori
Reply all
Reply to author
Forward
0 new messages