IVADO Presentation
Last friday, on October 22nd 2020, was held the third part of a virtual meeting by IVADO entitled "Digital October". This session was dedicated to "Tommorow's algorithm" with 6 student's presentations [1]. This conference was targetting a fairly large audience with students, researchers and industrialists, gathered around issues related to the use of data.
In that context, I had the opportunity to introduce a part of my research topic : an extension of Physics Informed Neural Networks with a modal approach. The idea was to describe the context, related challenges and fields of applications of this kind of techniques, rather than the actual results. I tried to show examples on how to use decomposition methods to make a problem easier to approximate so that it is possible to address more complex configurations.
This was also a new experience as we were experimentating a live conference on a lightboard at Polytechnique Montréal. I was filmed behind a large glass and my slides (with a black background removed) were added to this capture using OBS studio. Then the video stream was sent to Zoom through a virtual camera channel. In the end it proved to work rather well (except some windows that poped up at random time, but it was more likely a problem from my computer !). With more training, there could be a better use of this set up to make the presentation more dynamic. A huge Thank to Frédérick and Sébastien for the time and help they gave me to prepare this talk.
The replay is available on the LM2 Youtube Channel [2] and more detailed results might be presented in an article currently under writing (with code and data) !
[1] IVADO website - Digital October page https://ivado.ca/en/events/octobre-numerique-ivado/
[2] An extension of Physics Informed Neural Networks with a modal approach – IVADO Octobre Numérique - on Youtube : youtube.com/watch?v=uJHcULXF1HE