In this interview, Executive Director Yves Jacquier talks about La Forge’s expertise in prototyping and research in collaboration with academics, and presents a recent project on climate change visualization in collaboration with MILA, a deep learning research center in Montreal. This project will raise awareness of the future consequences of climate change with visual projections generated by solid scientific data.
Yves Jacquier oversees innovation in some production departments at Ubisoft Montreal. Four years ago, he has helped Ubisoft take ownership of new technologies by creating La Forge, a space where external academics and Ubisoft team members gather to work on research projects and prototyping.
In early March, Yves was meant to attend UNESCO’s Mobile Learning Week, a United Nations flagship event on Information and Communication Technology in education. While the UNESCO event was cancelled due to the evolution of the COVID-19 pandemic, the project is moving forward and the data set produced by Ubisoft La Forge is now available publicly on GitHub.
The Ubisoft Daily team reached out to Yves to learn more about his role at Ubisoft, the La Forge model, and how Ubisoft participated in the Visualizing climate change project.
Ubisoft Daily: First of all, could you tell us a bit more about yourself and your role at Ubisoft?
Yves Jacquier: I started my career in particle physics at CERN (European Organization for Nuclear Research). I also worked in medical instrumentation, telecoms, and even music. I joined Ubisoft Montreal 16 years ago, and am currently heading the production services departments, which essentially consists of guiding innovation in these departments.
For example, to produce MOCAP data, our production pipelines must be as efficient as possible, but we need also to keep exploring and test, integrate, and improve them to be ready for future productions. We’re always thinking of the future, making sure that we’ll still be leaders in cutting edge technologies in five years. As a result, my responsibilities have naturally evolved to help Ubisoft take ownership of new technologies such as telemetry, biometry, streaming, performance capture and now of course Machine Learning.
UD: Could you describe La Forge’s mission?
YJ: La Forge was created four years ago to accelerate research & development at Ubisoft. It’s a place where external academic researchers and Ubisoft team members work together on high potential prototypes that serve both interests: for us, improve our games, for them, enable scientific publications. We were the first in the gaming industry to develop such an R&D recipe where researchers have access to all of Ubisoft’s technologies, including game engines, data, as well as the expertise of our specialists. In exchange, we get access to and the chance to participate in high-level academic research.
La Forge was created as an interdisciplinary space by design. And because of that, we discovered that not only were we able to work together to improve our games, while creating public knowledge, we also contribute to solve real world problems.
UD: Ubisoft La Forge recently contributed to a project led by Mila (Montreal’s Institute for Artificial Intelligence) revolving around climate change awareness. Can you describe what this project is about?
YJ: This project is led by Yoshua Bengio, a well-known figure in AI, especially for his work on deep learning, and a long-term partner of La Forge. The Visualizing climate change project aims to make climate change more concrete for everyone by using Google Street View to generate “flooded” version of an individual’s address. The concept is that you can enter your address and, if you live in a floodable zone, the application will generate a flooded image of your home and the street where you live in 2050 using sound scientific climate models. You can see it as a time machine that is able to show you an accurate version of your neighborhood in 2050.
UD: What did Ubisoft contribute to his project?
YJ: To generate the images, the AI system needed to be trained by being presented with examples of flooded and non-flooded scenes, until it could generalize a set of rules that could help it to predict what certain features look like when the level of the water rises.
One of the issues was the lack of examples to train such models, since luckily there are not many real-life examples of non-flooded areas that suffered flood damage over the years. We used assets from the version of the San Francisco Bay Area in Watch Dogs 2, generating flooded and non-flooded versions of certain places in the game engine to feed more images to the AI and help it learn.
One of the questions we had on our end was whether the images generated in the game engine would be believable enough to train the AI model. What we found was that, when the AI had learned by using a combination of real-life images and our in-game images, it produced results that were three times more credible. It also allowed us to explore the limits of the realism in our game engine.
UD: In your opinion, how could this project eventually be used for educational purposes?
YJ: We humans have a cognitive bias that makes it difficult for us to make decisions when asked about long-term impacts, since they seem very abstract. Therefore, if we want to educate people about climate change and make them conscious of what they can do to avoid it, we must show them concrete examples they can relate to. The idea with this first project is to show the consequences of flooding, but eventually, the underlying technology could be used to generate simulations involving other risks of climate change.
UD: What are the next steps for this project?
YJ: The first phase ended with the release of the simulated flood data set available publicly on our Github. Researchers can use the data we’ve provided to improve the project’s algorithm to train new AI to create even more believable images. As far as I can tell, it’s the first time that Ubisoft makes this kind of data set open source. These are assets from a video game, from a product we sell, so it’s a pretty big deal to share them publicly and it shows our commitment to research in general and this project in particular. We are now thinking about other ways we can participate and other types of data that we can provide for similar initiatives in the future.
As for Mila’s Visualizing climate change initiative, the results are not yet open to non-participants at the moment as they’re still refining the AI. However, everyone can participate by sending in images of flooded houses and streets to continue training their climate model. We learned a lot about our own technology participating in this project so I can’t wait to see where they take it next.
On our end, I’m happy to disclose that we recently hired Vahe, the student we worked with, and already committed a fraction of his time to support the project in the next year.