Back

Speaker "Emad Barsoum" Details Back

 

Topic

 Overview of GAN based training and its applications.

Abstract

Unsupervised and self-supervised training are gaining a lot of attention recently due to advances in new algorithms and the availability of large-scale unlabeled dataset. Labeling large dataset is expensive and error prone, be able to generalize model accuracy from a small labeled dataset with the help of a large corpus of unlabeled data becoming crucial and have been getting a lot of attentions lately.
 
One of the main goals of self-supervised training is to learn a rich presentation on auxiliary tasks and use such presentation to improve the primarily task. The idea is from auxiliary tasks the model can learn the dynamic underlying the system being trained and transfer such learning to the primarily objective.
 
Another important aspect of unsupervised training is learning the underlying probability distribution of the data. This enable us to generate a variation of the data that has the same probability distribution as the original dataset. There are important scenarios in which this is useful such as data augmentation, improving the quality of the data, filling gaps…etc.
 
In this talk, we will provide overview of various GAN techniques and algorithm target computer vision problems and talk about latest trends in unsupervised training.

Profile

Emad Barsoum is an Architect at Microsoft AI Platform team. He leads the deep learning framework effort at Microsoft and help driving Microsoft strategy in AI. Prior to that Emad was Principal SDE and Applied Researcher in the Advance Technology Group at Microsoft Research. He was one of the core developer and researcher behind the Emotion Recognition algorithm used in MS Cognitive Service for both still image and video. Before that, He was one of the main Architects for NUI API on Xbox One, and the tech lead for the depth reconstruction pipeline for Kinect v2. His current research focuses are in computer vision and deep learning algorithms, especially in the area of activity detection/recognition and unsupervised learning. He has given numerous internal and external talks on Deep Learning and Computer Vision. He received his M.S. degree from U.C.Irvine and his doctoral degree from Columbia University.