Maciej Zięba & Tomasz Trzciński (Tooploox): Representation learning with and without GANs

Industry Talk at the PL in ML: Polish View on Machine Learning 2018 Conference (plinml.mimuw.edu.pl).
Abstract:
Learning a compact and efficient representation of images, words or 3D shapes is critical to many real-life applications, including visual search, 3D reconstruction and object recognition. In this talk, we will present several methods developed and published at top-tier conferences by Tooploox AI team members, starting from binary image descriptors learned with linear projections to unsupervised training of 3D point cloud representations built with Generative Adversarial Networks to be applied in the context of autonomous cars.