We work with the community to systematize and speed up complex, ad-hoc and error-prone performance/energy/size/cost/reliability optimization and co-design of computer systems using our customizable repository of knowledge, auto-tuning, run-time adaptation, big data, predictive analytics, crowdsourcing, collective intelligence and reproducible research.

News and upcoming events:

Grigori Fursin, PhD

Generated by Collective Mind V1.0

Grigori Fursin was one of the first researchers to radically change ad-hoc, error prone, time consuming and costly process of benchmarking, optimizing and co-designing computer systems across all software and hardware layers (heterogeneous multi-core architectures, compilers, run-time libraries, applications) for minimal execution time, power consumption, failures and other costs into a unified big data problem [P23, P1, M2, P2, P14, P16, P5]. He then started tackling, systematizing and speeding it up using his public cTuning.org and Collective Mind repository of knowledge, common plugin-based application and compiler auto-tuning infrastructure, statistical analysis, machine learning (classification and predictive analytics), data mining (finding missing features), run-time adaptation, brain-inspired self-optimization, crowdsourcing using even commodity mobile phones, social networking and collective intelligence. As a side effect, Grigori's research initiated new publication model in computer engineering where experimental results and all research artifacts are continuously shared, discussed, reproduced and improved by the community [E1,E2,P2]. This research was validated in industry, received several international and national awards, and helped to improve utilization of computer systems while dramatically reducing their development costs and time to market. We hope it will help improve our society by boosting innovation in science and technology.
Current position(s):
Main achievements and awards:
  • 2012-2016: Received INRIA award and fellowship for "making an outstanding contribution to research".
  • 2010-2011: On industrial leave invited to help establish new Intel Exascale Lab in France based on cTuning technology, serve as a head of application characterization and optimization group, and direct research and development.
  • 2008-cur.: Established the nonprofit cTuning foundation serving as a President and CTO.
  • 2007-cur.: Regularly transferring developed technology to industry and production tools such as mainline GCC; consulting major companies on systematic and reproducible program and architecture performance tuning, run-time adaptation and co-design.
  • 2006-2009: Led research and development of the machine-learning based self-tuning compiler (Grigori proposed to combine predictive analytics, big data and collective intelligence) in EU FP6 MILEPOST project considered by IBM to be the first in the world .
  • 1999-2006: Prepared foundations for big data driven and machine learning based optimization, run-time adaptation and co-design of computer systems.
  • 1993-1998: Designed novel semiconductor neural network accelerators for a possible brain-inspired computer.
Main interests and expertise: Research and development:
  • enabling self-tuning computer systems that can automatically adapt all their software and hardware layers at can automatically adapt to any user task for minimal execution time, power consumption, failures and other costs
  • developing collaborative, multi-objective optimization and dynamic adaptation using big data analytics (statistical analysis, data mining, machine learning)
  • enabling reproducible research in computer engineering
  • developing public repositories of knowledge (semantically or directly connected code and data) and collaborative experimental infrastructure
  • investigating biologically and brain-inspired systems, neural networks, artificial intelligence, physics, electronics

Management:

  • preparing and leading challenging, long term, interdisciplinary R&D projects
  • building and leading teams of researchers and developers

Transfer to industry:

  • consulting companies on systematic, machine-learning based and reproducible program and architecture analysis, benchmarking, optimization and co-design
  • moving technology to production design and optimization tool chains including open source GCC and LLVM compilers
  • setting up joint industrial and academic laboratories
Education:
Languages: English, Russian, French basics
Professional memberships: ACM, HiPEAC, IEEE
Full academic CV: HTML; PDF
LinkedIn: Link
Research twitter: Link
EMail: grigori.fursin@inria.fr  or  grigori.fursin@cTuning.org

Quick access to cM-generated Curriculum Vitae:

Institution building (I) ],
Keynotes (K) ],
Startups (C) ],
Examiner ],
Expert service (E) ],
Major research achievements (M) ],
Public or in-house repositories of knowledge (R) ],
Awards, prizes, and fellowships (A) ],
Major funding (F) ],
Professional experience (J) ],
Education (Z) ],
Major software and datasets (S) ],
Hardware (H) ],
Talks (T) ],
Participating in program committees and reviewing ],
Teaching and organizing courses (L) ],
Advising/collaborating (Q) ],
Organizing/chairing events (E) ],
Publications (P) ]
Miscellaneous ]

Grigori Fursin is a computer scientist, software developer, consultant and entrepreneur with an interdisciplinary background in physics, electronics, computer engineering, and machine learning and with more than 20 years of research and development experience. He is always interested to consult companies and lead highly challenging, innovative and interdisciplinary research and development projects particularly related to development of faster, smaller, cheaper, more power efficient and reliable self-tuning computer systems; collaborative discovery, systematization, sharing and reuse of knowledge; big data; predictive analytics; electronic brain; artificial intelligence; bio-inspired devices; systematization of benchmarking; automation of compiler tuning; robotics; space exploration; semantic web; crowdsourcing using even commodity mobile phones, social networking and collective intelligence.

In 1993, at the age of 16, Grigori joined his first official R&D project as a Research Assistant designing and optimizing semiconductor neural network accelerators for a possible brain-inspired supercomputer computer [M9, M8]. After amateur and tedious attempts to optimize and parallelize his neural network modeling software for several supercomputers, Grigori decided to switch to computer engineering to automate and systematize this process using his interdisciplinary knowledge. Eventually, Grigori was one of the first researchers to radically change ad-hoc, error prone, time consuming and costly process of designing, benchmarking and optimizing the next generation of computer systems across all software and hardware layers into a unified big data problem [P1, M2, P23, P14, P16, P5]. He then started tackling, systematizing and speeding it up using his public cTuning.org and Collective Mind repository of knowledge, common plugin-based auto-tuning infrastructure, statistical analysis, machine learning (classification and predictive analytics), data mining (finding missing features), run-time adaptation with static multi-versioning, adaptive exploration of large optimization spaces, online tuning, differential analysis, brain-inspired self-optimization, crowdsourcing using even commodity mobile phones, social networking and collective intelligence.

Besides publishing in major conferences and journals including PLDI, MICRO, CGO, TACO, IJPP, CASES and HiPEAC, Grigori spends considerable effort to release all his artifacts including tools, benchmarks, data sets and predictive models along with his articles at cTuning.org and c-mind.org/repo to ensure reproducibility. As a side effect, this initiated new open publication model in computer engineering where experimental results and all research artifacts are continuously shared, discussed, reproduced and improved by the community [E1,E2,P2, M1,E9,P5].
This also helped to move Grigori's techniques to mainstream production environments including GCC while validating and extending them in industry with IBM, ARC (Synopsys), Intel, Google, STMicroelectronics and ARM. These techniques enabled practical open-source machine learning based self-tuning compiler (MILEPOST GCC) [M3,R3,R1] considered by IBM to be the first in the world [P22]. They also dramatically reduced development costs of the new embedded reconfigurable devices from ARC (Synopsys) while improving time to market and ROI [X8]. Finally, Grigori's techniques were included in the EU HiPEAC IT research vision for 2012-2020 [X7, X2] In 2008, Grigori established international nonprofit cTuning.org foundation to help academic and industrial partners systematize, automate and speed up design, benchmarking optimization of their computer systems (software and hardware stack) using his open source repository, tools and big data analytics.

Grigori delivered more than 60 regular and invited talks, lectures and keynotes in major international companies and universities in Europe, USA, China, Canada, Russia and Taiwan; founded SMART and ADAPT workshops that ran consecutively for 8 years sponsored by Google, NVidia, Intel and Microsoft; prepared and taught advanced MS course on future self-tuning computer systems in Paris South University. In 2010-2011, he was on industrial leave invited to help establish Intel Exascale Lab in France while preparing long-term research directions and serving as the head of program optimization and characterization group [I1]. In 2012, Grigori rejoined INRIA and received a 4-year fellowship for "making an outstanding contribution to research" [A1]. Grigori hopes his research will help improve our society by boosting innovation in science and technology.

Current research focus

  • Continue preparing and promoting new publication model in computer engineering to validate experimental results by the community [E1,E2,M1,E9,P5,S1,R1].
  • Systematize and automate code and architecture co-design and optimization within public Collective Mind Infrastructure and Repository (plugin-based knowledge management system) based on multidisciplinary approaches including empirical tuning, machine learning, statistical analysis, and crowdsourcing using even commodity mobile phones [S1,R1]
  • Analyze, improve and compact hybrid predictive models and associated sets of canonical program, dataset and architecture features
  • Consolidate all past research and developments for French HDR
  • Transfer knowledge to industry through consulting (development of fast and power-efficient, intelligent and self-tuning computer systems; intelligent financial services; space exploration; pharmaceutical industries; brain modeling, etc) and creation of interdisciplinary laboratories

Keywords: Collective Mind, program and architecture crowdtuning, plugin-based auto-tuning, machine learning, data mining, public repository of knowledge, big data, crowdsourcing, compilers, online tuning, run-time adaptation, software and hardware co-design, electronic brain, neural networks, predictive analytics, statistical analysis, feature analysis, decremental analysis, decremental characterization, complexity reduction, reproducible research, crowdsourcing experimentation, cTuning.org, c-mind.org, agile research and development, knowledge systematization, emerging information technologies, knowledge transfer to industry, consulting, startups


Miscellaneous