The Speakers
Invited Speaker: Associate Professor Truyen Tran
Truyen Tran is Associate Professor at Deakin University where he leads a research team on the next generation of deep learning and applications to computer vision, computational science, biomedicine and software analytics. He publishes regularly at top AI/ML/KDD venues such as NeurIPS, ICML, ICLR, CVPR, UAI, AAAI, IJCAI and KDD. Tran has received multiple recognitions, awards and prizes including Best Paper Runner Up at UAI (2009), Geelong Tech Award (2013), CRESP Best Paper of the Year (2014), Third Prize on Kaggle Galaxy-Zoo Challenge (2014), Title of Kaggle Master (2014), Best Student Papers Runner Up at PAKDD (2015) and ADMA (2016), and Distinguished Paper at ACM SIGSOFT (2015). He obtained a Bachelor of Science from University of Melbourne and a PhD in Computer Science from Curtin University in 2001 and 2008, respectively.
Title: Robust visual reasoning
Abstract: Visual reasoning is the process of deliberatively inferring a conclusion from a visual content in response to a query. This is typically recognized in a system that integrates fast visual recognition and slow multi-step relational predicate chaining and refinement. Systems that learn to reason from demonstrated data suffer from not only the same vulnerabilities of typical machine learning systems due to external adversaries but also the lack of robustness intrinsic to reasoning. In this talk, I will discuss our research on tackling three main issues: Trojan attacks on visual recognition, lack of logical consistency in the answering, and lack of meaningful object and concept selection in the reasoning process. I will present effective solutions to these problems, including Trojan filtering, introducing logical constraints, and acquiring and applying attention priors from data to guide the reasoning process.