�Ӱ�ֱ��

Abstract: Structural graph theory is making great advances in understanding what characteristics of graphs act to make those graphs difficult or easy for algorithms for a range of common problem types (maximum clique, graph cuts, etc.). Modern machine learning is largely driven by the success (or otherwise) of a proposed model/algorithm on standard test datasets. This involves a plethora of different measures and sometimes one might suspect that a dataset has “easy examples” and “hard examples” and that a statistic might be affected by the proportion of hard or easy samples. So a structural graph theory based understanding of AI problems should help both understand benchmark results and also to design algorithms according to the range of complexities of data a problem setting can “throw up”.

In this talk, Professor David Suter will sketch these ideas with reference primarily to robust model fitting (using Maximum consensus - MaxCon) but also arguing that more generally clustering approaches could also be addressed in a similar fashion. This talk is essentially about the ARC funded DP25 project "HyperGraph Classes, Robust Fitting and Clustering” and will outline some preliminary steps (restricted to the graph cases, essentially because that is simpler to explain, more is known, and the limits of a short talk - however, the main challenge is in extending these ideas to hyper graphs).

Biography: David Suter holds the following degrees: BSc (Applied Maths and Physics, The Flinders �Ӱ�ֱ�� of SA, 1977); Grad. Dip. Ed. (Secondary Teaching, The Flinders �Ӱ�ֱ�� of SA, 1978); Grad. Dip. Comp. (Royal Melbourne Institute of Technology, 1984); PhD (Computer Vision, La Trobe �Ӱ�ֱ��, 1991).

He has held the following appointments: Lecturer at La Trobe �Ӱ�ֱ�� 1988-1992); Senior Lecturer, Associate Professor and Professor at Monash �Ӱ�ֱ�� (1992–2008), Professor at The �Ӱ�ֱ�� of Adelaide (2008–2017).

Currently he is a research professor at Edith Cowan �Ӱ�ֱ��, Perth, �Ӱ�ֱ��.

He has served on the �Ӱ�ֱ��n Research Council College of Experts (2008–10) and the editorial boards of several journals, including: International Journal of
Computer Vision (2004–2013), Journal of Mathematical Imaging and Vision (200—2010), Machine Vision and Applications (2006–2008).