Yoshua Bengio

TL;DR: In this article, a broad meta-learning framework is proposed to discover generic ways of processing time series (TS) from a diverse dataset so as to greatly improve generalization on new TS coming from different datasets.

TL;DR: GibbsNet as mentioned in this paper uses an adversarially learned iterative procedure to refine the joint distribution to better match with the data distribution on each step, achieving the expressiveness and flexibility of an undirected latent variable model.

TL;DR: The authors proposed a bidirectional Helmholtz machine, which guarantees that the top-down and bottom-up distributions can efficiently invert each other by interpreting both the topdown and the bottomup directed models as approximate inference distributions and defining the model distribution to be the geometric mean of these two.

TL;DR: This work compares two recently-introduced state-of-the-art machine learning algorithms, Support Vector Machines and Discriminative Restricted Boltzmann Machines, and develops how to use them to solve this difficult acoustic classification task, and obtains classification accuracy results that could make these techniques suitable for fielding on autonomous devices.

TL;DR: OC-GAN as mentioned in this paper learns representations of the spatial relationships between objects in the scene, which leads to an improved layout-fidelity and improves the object instance-awareness of GANs.