The Power and Limits of Machine Learning Models

Date: 14 October 2020 (Wednesday)
Time: 6-7pm
Venue: Online via Zoom Meeting
Meeting ID: 964 2448 0558
Passcode: 8y567c
Title: The Power and Limits of Machine Learning Models (CSE Festival UG seminar)
Speaker: Prof. Dit-Yan Yeung, CSE

Introduction:

Machine learning has achieved unprecedented success in many applications over the past few decades, especially in the most recent 8-10 years as a result of significant advances in deep neural networks. Consequently, real-world applications of machine learning are commonly seen nowadays. To appreciate the power of machine learning models, we will start with taking a quick look at some splendid machine learning applications developed in recent years. In the second part of the talk, the focus will be shifted to discussing the robustness of existing machine learning models, particularly deep neural networks. Among other recent findings, it has been demonstrated that many well-trained machine learning models which deliver high prediction accuracy can be fooled easily using carefully manipulated examples to result in incorrect predictions that differ significantly from human judgment. These recent findings open up interesting research directions that span both the machine learning and cybersecurity areas.

Speaker:

Prof. Dit-Yan Yeung is a Chair Professor and the Head of the Department of Computer Science and Engineering at HKUST. Since he was a research student, his primary research interests have been in computational and statistical approaches to machine learning and artificial intelligence. He is also interested in novel application of machine learning to many different domains, including computer vision, education, and urban computing. In addition to conducting research, he is passionate about bringing artificial intelligence education to a broader audience to demystify this popular area amid a storm of hype.

Download presentation slides.