Skip to main content
Workshops & seminars, Conferences & lectures

Seminar by Dr. Yani Ioannou (University of Toronto)


Date & time
Friday, April 9, 2021
10 a.m. – 12 p.m.
Speaker(s)

Dr. Yani Ioannou

Cost

This event is free

Where

Online

Title: Efficient Deep Learning

Abstract:
Deep Neural Networks (DNNs) are behind many of the most exciting research frontiers and applications of AI in fields such as computer vision and natural language processing. With much of this progress having been enabled by an increase in computational resources and hardware acceleration, future research and application of the next generation of DNNs, including Transformers (such as GPT-3), and large semi-supervised models (such as SimCLR), are currently limited by the massive computational requirements of both training and using these state-of-the-art DNNs. In this talk, we will cover some of my PhD work on Convolutional Neural Networks with filter groups and low-rank filters, methods of structured sparsity now commonly used to improve both the efficiency and generalization of DNNs for computer vision, and used in products such as Apple's Face ID, and Google Photos. We will then delve deeper into my recent work at Google Brain on training sparse neural networks and, finally, we will discuss the need for efficient deep learning in the next generation deep learning architectures, and how better understanding sparse neural network training could help both the future of AI research, and edge AI applications.

Bio:
Yani Ioannou is currently a sessional lecturer at the University of Toronto, where he teaches APS360: "Applied Fundamentals of Machine Learning" in the Faculty of Applied Science & Engineering. Until recently, he was a Visiting Researcher at Google Research in Toronto where he focused on sparse neural networks, and previously a Research Scientist at a UK self-driving startup, Wayve. Yani obtained his PhD at the University of Cambridge in the Department of Engineering under the supervision of Professor Roberto Cipolla, head of the Computer Vision and Robotics Group, and Dr. Antonio Criminisi at Microsoft Research. Yani is a recipient of a Microsoft Research PhD Scholarship and collaborated with Microsoft Research Cambridge throughout his PhD, where he developed impactful efficient deep learning methods, including the first paper to propose using filter groups to improve representations learned in Convolutional Neural Networks. Outside of deep learning research, Yani has helped NASA with exoplanet detection, the Gates Foundation with malaria classification, and contributed to open source projects such as the Linux kernel and the Point Cloud Library.

Back to top

© Concordia University