John Canny
John Canny received his B.Sc. in Computer Science and Theoretical Physics from Adelaide University in South Australia, 1979, a B.E. (Hons) in Electrical Engineering, Adelaide University, 1980, an M.S. and a Ph.D. from M.I.T, 1983 and 1987, respectively. He joined the U.C. Berkeley faculty of EECS in 1987.
John Canny has been awarded an international chair by the Inria Chairs Committee for the period 2013-2017 within the WILLOW research team in Paris-Rocquencourt on the topic “Active Robotic Agents for Preschool Learning”.
John’s field of research is Big Data analytics and applications, especially in health care and education.
– John, can you tell us about the team you work with? For how long have you been working together? On what project are you currently working on together?
With the WILLOW group in Inria Paris-Rocquencourt we are working with on a project to improve literacy among disadvantaged children. These are the kids in the lowest SES (Socio-Economic Scale) end of the population. In the US, they have a 2-3 fold deficit in language comprehension and ability to use language, relative to other children. This creates a handicap when they reach school which slows their learning, leading to a cycle of lagging behind which persists throughout school and beyond. Children from more privileged home receive constant language practice throughout their day. We’re trying to recreate that practice through appropriate technology that could be embedded in an appropriate way – the kind of cheap but smart robots that you can find in toy stores now. Using computer vision, deep learning and simple language understanding, we can build interactive games that should engage the child in using language and learning about the world. Thanks to recent progress in image and video understanding and language generation, especially at WILLOW, this system is technically possible although still very challenging. On our end we have been developing tools for embedded data analytics and also working with children on similar language games.
– Why did you apply for the Inria International Chair position and what does it bring you?
Through prior interactions with WILLOW, I learned that the Chair program was being created. The chair gives me the opportunity to work regularly with the WILLOW team, one of the great vision groups in the world, and the flexibility to tackle this ambitious project in a long time frame. The Inria groups are physically close together, which makes it easy to engage with the other machine learning researchers at Inria. And of course it’s a fantastic experience for myself and the family to stay in Paris over the summer. Everywhere you look, it has a sense of place and history.
– How is your time in France organized?
Most of the time is at Inria in Paris. That’s my home base. It also allows regular interaction with Ecole Normale Superieure nearby which has a strong group also in computer vision. I visited with Cordelia Schmid’s group in computer vision (Grenoble) last summer and hope to visit PIerre-Yves Oudeyer’s group in cognitive robotics (Talence) this summer.
– What do you expect from this collaboration and what would be the next step?
Right now, we’re developing basic algorithms for image understanding and language generation. That’s going to continue for a year or two. This work is right at the bleeding edge of computer vision right now, and it’s important to allow it to mature a bit. After that we would like to move to prototypes and studies in an experimental preschool associated with UC Berkeley. Beyond the chair, we hope to continue the collaboration through the Inria@SiliconValley program.
– According to you, what are the main differences between France and the US regarding research?
In the US, we have a strong tradition of industry research – although there has been some contraction in industry research in recent years, it is still very significant on academic publication, especially in areas like deep learning. I’m currently working half-time at Yahoo research labs and half-time at Berkeley. This provides a great window into real-world data analysis challenges, and opportunities to study the practice of data analysis in the wild. It also fosters student/company interactions which are extremely helpful for those students who will seek an industry job after graduation.
There are also many mechanisms to support the migration of technology into industry in the Bay Area, i.e. through startup incubators. Berkeley has several program including the Skydeck incubator space, and there are many, many meetup programs in the SF Area.
In France and Europe more generally there seems to be an emphasis on longer, more strategic research projects. I think that’s very good for tackling large challenges that are beyond the means of a single faculty member or even a group.
– What advice would you give to a young American researcher who is willing to work with a French Team?
There are several possibilities. The easiest would probably be to find a US Professor who already collaborates with a French team. There are quite a few of those at Berkeley and Stanford. They will likely have resources for travel and the personal connection to recommend you to the French team.
The next best choice we be to apply for funds, either through an exchange program like Fulbright or through smaller exchange programs. Inria@SiliconValley has several of these for post-docs and graduate students.
Some creativity in finding funds will often be needed. Stay with it. It really is a worthwhile experience (I visited France twice for longer visits as a graduate student). France is obviously a beautiful place, but it is also very different world view from what you experience in the US. It is a mature culture, and has survived through centuries, building art, literature and tradition all the time.
Interviewed by Tania Castro, European & International Partnerships Department, Inria
