Grigori Fursin

Playground    Mission    LinkedIn    Google scholar    Twitter    Medium    GitHub    Discord    EMail



My latest news:

2024 March: Arjun Suresh and I will present our new project to "Automatically Compose High-Performance and Cost-Efficient AI Systems with MLCommons' Collective Mind and MLPerf" at the MLPerf-Bench workshop @ HPCA'24.

2024 January: my cKnowledge Ltd is now working for MLCommons to enhance reusable Collective Mind automation recipes for MLOPs and DevOps to make it easier for the community to build, benchmark and optimize AI systems across rapidly evolving models, software and hardare. It is a community project based on feedback from many great colleagues from Google, AMD, Neural Magic, Nvidia, Qualcomm, Dell, Nutanix, OctoML, HPE, Red Hat, Intel, TTA, One Stop Systems and other organizations. Our current goal is to provide a single common interface to run and reproduce MLPerf inference benchmarks v3.1 and v4.0 on any software and hardware stack from Intel, Nvidia, Qualcomm, AMD and other vendors in a unified way either natively or inside automatically-generated containers - please join Discord server to participate in this project and get free help with your MLPerf inference v4.0 submissions.

2023 September: the cTuning foundation and cKnowledge Ltd are proud to deliver the new version of the MLCommons CM workflow automation language, CK playground and modular inference library (MIL) that became the 1st open-source technology enabling mass submission of 12K+ performance results in a single MLPerf inference submission round with more than 1900 power results across more than 120 different system configurations from different vendors (different implementations, all reference models and support for DeepSparse Zoo, Hugging Face Hub and BERT pruners from the NeurIPS paper, main frameworks and diverse software/hardware stacks) in both open and closed divisions - see related HPC Wire article for more details about our CK and CM technology.


You can learn more about my recent community initiatives, automation tools, platforms and industrial projects with MLCommons, IBM, Intel, General Motors, Arm, Google, Amazon, OctoML, HiPEAC, ACM, IEEE and other great collaborators from my ACM TechTalk'21, keynote at ACM REP'23, overview in Philosophical Transactions of the Royal Society'21 and my reproducibility initiatives at ML and Systems conferences since 2014. Feel free to reach me via our Discord server and connect at LinkedIn.


My current activities:
  • author of the open-source, non-intrusive and technology-agnostic Collective Mind workflow automation language (CM) adopted by MLCommons (125+ AI organizations) to make it easier to benchmark and optimize complex AI/ML systems across rapidly evolving models, data sets, software and hardware from different vendors (Intel, Nvidia, Qualcomm, AMD ...);
  • founder of the Collective Knowledge Playground to optimize AI systems via community challenges (trade off performance, power consumption, accuracy, cost and other characteristics) powered by MLCommons CM automation;
  • organizer of reproducibility initiatives and artifact evaluation at ML and Systems conferences in collaboration with ACM, IEEE and MLCommons since 2013 - we develop a common common methodology and interface to share, manage, run and reuse code, data and experiments from published papers. See my ACM Tech Talk'21 and ACM REP'23 keynote for more details.
  • co-chair of the MLCommons Task Force on Automation and Reproducibility extending CM workflows to modularize MLPerf benchmarks and run them in a unified way on any platform with any SW/HW stack based on feedback from the research community, Google, AMD, Neural Magic, Nvidia, Qualcomm, Dell, HPE, Red Hat, Intel, TTA, One Stop Systems and other organizations.
  • founder and president of cTuning.org (non-profit organization) and co-chair of the MLCommons Task Force on automation and reproducibility developing open-source tools that faciliate reproducible research and bridge the growing gap between AI research and production.
  • founder of cKnowledge Ltd working with MLCommons and other companies to develop the Collective Mind workflow automation language and Collective Knowledge Playground - our goal is to make AI accessible to everyone by solving the growing complexity of AI systems, automating the development and deployment of highly-efficient AI/ML systems, and reducing all operational costs.
  • consultant for startups and investors to help them avoid numerous pitfalls and minimize risks when developing and deploying state-of-the-art technology based on AI.
  • angel investor and consultant for open-source, open-science and educational startups to make AI accessible to everyone.
My past activities:
  • author of the Collective Knowledge technology v1 and v2 adopted by MLCommons (125+ AI organizations);
  • vice president of MLOps at OctoML;
  • founder and chief architect of cKnowledge.io acquired by OctoML;
  • founder in residence at Enterpreneur First;
  • co-director of Intel Exascale Lab and tech.lead for performance analysis and optimization;
  • senior tenured scientist at INRIA;
  • research associate at the University of Edinburgh;
  • PhD in computer science from the University of Edinburgh with the Overseas Research Student Award;
  • author of the Artifact Evaluation and Reproducibility checklist for ACM conferences;
  • recipient of the European technology transfer award, ACM CGO test of time award and INRIA award of scientific excellence for my original research to use AI to automate development of efficient AI systems, and for the development of a self-optimizing ML-based compiler considered to be the 1st in the world.
My brief biography:

I am a computer scientist, software engineer, business executive, angel investor, educator, lifelong learner, and adventurer with more than 15 years of professional experience leading innovative projects to make AI fast, efficient and accessible to everyone in collaboration with MLCommons, IBM, Intel, General Motors, Arm, Amazon, Google, HiPEAC, ACM, IEEE, and other organizations.

My passion is to work with the community to solve real-world problems, It turned out that I was among the first researchers to use AI to automate the development of efficient software and hardware while open-sourcing all my code, data, models, automation workflows, experimental results, tech. reports and other artifacts in 2008 when it was still a taboo. Even though it was a very painful experience to go against the current and try to change the Status Quo in ML and Systems research back then, I did that because I strongly believed in the power of open science, open source, collaboration, and knowledge sharing to accelerate innovation and solve the most challenging problems.

I am very glad that this effort was not wasted and eventually helped establish reproducibility initiatives and artifact evaluation at practically all ML and Systems conferences. It also allowed me to esablish cTuning.org and cKnowledge.org and continue collaborating with the community and many companies, startups, and non-profits including MLCommons, IBM, Intel, General Motors, Arm, Amazon, Google, INRIA, HiPEAC, IEEE and ACM to develop open-source automation tools and platforms including Collective Knowledge (CK) and Collective Mind (CM) that I donated to MLCommons in 2022 to benefit everyone.

I continue leading these community projects to solve the growing complexity of software projects and AI systems; make it easier to prototype, validate and reproduce research ideas across rapidly evolving models, data, software, and hardware; automate the development of fast and efficient computer systems using AI; accelerate innovation and technology transfer; and make AI accessible to everyone in an open and fair way.

At this moment, I focus on the development of the open-source Collective Mind workflow automation language and Collective Knowledge Playground with the community and MLCommons (125+ AI organizations) to bridge the growing gap between AI research and production, facilitate reproducible research, and accelerate AI innovation. Our goal is to help everyone benchmark and optimize numerous AI/ML models across continuously changing data, software and hardware, and validate them in the real world across diverse and rapidly evolving AI/ML models, data, software and hardware from the cloud to the edge.

In my spare time, I continue helping ML and Systems conferences organize and automate reproducibility initiatives. I also use my experience to help investors and startups avoid numerous pitfalls and reduce risks and costs when bringing complex ML/AI systems to production via my cKnowledge Ltd. Finally, I have a few stealth projects that should go live around mid 2024.

When I have time and resources, I am glad to give invited talks and lectures, and help new projects and initiatives related to automating and simplifying development and deployment of efficient AI systems, supporting reproducible research, enabling open-science, and making AI fast, efficient and accessible to everyone - you can reach me via Discord server or email!

I am very grateful to my fantastic colleagues and collaborators who helped develop Collective Knowledge framework (2014-2022) and Collective Mind technology (2022-cur.)!

You can find more details about my R&D timeline, professional career and projects in the extended version of this page.