Basser Seminar Series

Machine Learning Unhinged

Speaker: Dr Brendan van Rooyen
Speaker QUT and UC Berkeley

When: Wednesday 13 April, 2016, 4:00-5:00pm

Where: The University of Sydney, School of IT Building, SIT Lecture Theatre (Room 123), Level 1

Add seminar to my diary

Abstract

The Support Vector Machine (SVM) and its hinged loss function provide powerful means to learn classifiers from data. While the practical utility of the SVM is undoubted, it does have drawbacks, chief amongst these its non-robustness to label-noise. In this talk I will summarize recent work I presented at the Conference on Neural Information Processing systems (NIPS for short) on the topic of learning from corrupted data, work that suggest means to make SVM’s robust to label noise. I will show that the SVM needs to be unhinged to realise its full potential.

Speaker's biography

Having recently relocated from Canberra and the Australian National University, where he completed a PhD on the topic of machine learning theory, Brendan van Rooyen now splits his time between QUT and UC Berkeley. He enjoys all the things decision theory, warm weather and writing about himself in the third person. More information.