Prof. Alex Smola’s lecture on Deep Learning on Graphs and Practical Computer Vision with GluonCV

Speaker : Prof. Alex Smola
Distinguished Scientist and Director for Machine Learning Amazon Web Services

Title : Talk 1: Deep Learning on Graphs (From – 3:30 PM to 4:15 PM)
Talk2: Practical Computer Vision with GluonCV (From – 4:30 PM to 5:15 PM)

Date : Thursday, May 16, 2019

Time : 3:30 PM

Venue : CSA Seminar Hall (Room No. 254, First Floor)

Abstract

Talk 1:
Deep Learning has shown great success on a wide range of domains, such as text, images, audio and tabular data. For structured data on graphs deep learning remains a work in progress. In this talk I’ll cover the basics of models with vertex updates and stationary distributions. Moreover, I’ll address how to distribute and partition the problem efficiently for scalable inference. Lastly, I’ll describe how DGL.ai offers a high-level Python API to accomplish this goal, thereby allowing the algorithm designer to focus on the model while a systems designer can focus on computational efficiency.

Talk 2:
Computer vision is a key tool in the modern deep learning toolbox. In this talk I will give a brief overview of the GluonCV open source toolkit for object recognition, detection, segmentation, and pose estimation. It allows users to get started easily with advanced models such as ResNext, Yolo, SSD, MaskRCNN or SimplePose. The talk includes live demonstration of the models and details on how to train them efficiently. Beyond computer vision I give a brief introduction to D2L.ai, a novel open source textbook on deep learning, based on Jupyter notebooks, that is fully executable by its readers.

Biography of the speaker

Alex Smola is currently a Distinguished Scientist and Director for machine learning at Amazon Web Services. Prior to joining Amazon he was a faculty at CMU, and NICTA. For more information about him please see
https://alex.smola.org/

Host Faculty : Prof. Chiranjib Bhattacharyya