Olivier Koch

Computer vision and deep learning in the real world

I lead the machine learning and MLOps teams at Onfido. We use computer vision and deep learning for online identity verification. From 2015 to 2021, I led the machine learning team for recommendation at Criteo. From 2010 to 2014, I led the computer vision team at Thales Optronics in France.

I graduated with a PhD at MIT in 2010 under the supervision of Prof. Seth Teller. I was a core member the MIT team at the DARPA Urban Challenge in 2007. My work was published at CVPR, ICCV, ICRA and IJFR.

Posts

also on Medium

Featured talks

  • 2023, June 4th: ML Prague, Bringing automation and fairness to identity verification on the internet with deep learning - slides
  • 2023, June 21st: An introduction to ChatGPT and generative AI, Point Nine CTO Meetup, Paris - slides
  • 2023, March 30th: ChatGPT: have we crossed the Rubicon? Invited talk at BPCE DigiPay Tech+ Event
  • 2022, IFTTD podcast: Large-scale AI (in french), youtube video
  • 2020, Job interview talk at Onfido, Scalable representation learning and retrieval for online advertising
  • 2018, 7th RecSys London Meetup, Acquisition: Recommending Products From a Retailer That the User Never Visited Before
  • 2017, Criteo RecSys London, Product recommendation beyond collaborative filtering, welcome to the twilight zone!
  • 2016, ICML: Online Advertising Systems Workshop, New challenges for scalable machine learning in online advertising
  • Oct 2014, ICIP, Image and Video Processing for Defense, Transportation, Homeland Security, and Observation from Space: Industrial Expectations and Technological Challenges

Publications

also on Google scholar

Teaching

  • Teaching the Deep Learning class at Master Level (3rd-year) at ENSAE, Spring 2024. Stay tuned!
  • Teacher at M2 level at Paris Dauphine University with Criteo, IASD, 2019-2020
  • Teacher at the Workshop on Recommendation, Ecole Polytechnique Data Science Summer School, June 2018 & 2019
  • Back in 2008, I taught Java at MIT over IAP. This was one of the first courses to end up on opencourseware the following years!

Misc

  • Book recommendations for tech leaders.
  • I co-founded Reveal, the RecSys workshop for causal recommendation with Microsoft, Netflix and Cornell University. The workshop ran from 2018 to 2020 and now continues as the Consequences workshop.
  • In 2015, I won the Criteo 2015 Hackathon (16 teams) with image-based similarity using deep networks: developed an end-to-end pipeline with Python/caffe/cuDNN/protobuf on AWS processing 400K images per hour and shipped to AB test
  • Google Hashcode: 5/65 in 2015 Finale Round, 76/1054 in 2016, 300/2815 in 2017