Hi! My name is Frédéric Branchaud-Charron, I’m currently an Applied research scientist at ElementAI and I did my MSc by Research at Université de Sherbrooke with Pr. Pierre-Marc Jodoin.
My research is focused on active Learning and uncertainty estimation. I also work on complexity measures, ie. being able to predict the achievable accuracy on a certain dataset without training a deep neural-network.
Open-Source
At ElementAI, I co-created BaaL, an active learning library.
I contribute to many projects such as Keras or keras-contrib.
I am member of the Keras-organization. My main contribution is the Sequence API and a total revamp of the input pipeline. I’m also project manager of keras-preprocessing.
I also built a new library keras-transform to perform flexible data augmentation.
On this blog, I’ll share many of my experiments using Keras and TensorFlow. I’m also using Pytorch for some projects.
Interests
- Active learning
- Uncertainty estimation
- Data pipeline
- GANs
- Functional programming (Elm, Haskell)
Academic work
- Publications
- MIO-TCD: A new benchmark dataset for vehicle classification and localization, Zhiming Luo, Frederic B.-Charron & al., Published in IEEE TIP 2018
- Spectral Metric for Dataset Complexity Assessment, Frederic Branchaud-Charron, Andrew Achkar, Pierre-Marc Jodoin, Published at CVPR 2019
- Other
- Reviewer MICCAI 2017
- Reviewer and co-organizer of TSWC 2018
Where to find me?
I’m currently working in Montréal at ElementAI in the active learning and uncertainty estimation team..
In my lab, University of Sherbrooke (D3-2003)
I’m currently doing an internship at Miovision as part of my MSc.
I’ll be at CVPR 2017 to present a new challenge at TSWC 2017!
I’ll be at Montreal AI Symposium 2017 to present a poster!
Contact me
-
Keras slack @dref360. Feel free to PM me!