Why is it so hard to get computers to count things? Examples from a Career in Applied Mathematics

Speaker: Paul Bellette 
Affiliation: Plotlogic

Abstract

Recent years have seen an explosion in applications of machine learning to various problems, with some incredible advancements on tasks that seemed out of reach only 15 years ago. The current prime examples probably being in generative models for images and text. However these systems often remain surprisingly bad at tasks that seem superficially trivial, such as counting. In this talk I will give some more modest examples that I have encountered in my career of where counting physical objects has been an important and yet surprisingly difficult task to get an automated system to perform. Along the way, this will also give a survey (or cautionary tale) of what a career in mathematics in Brisbane can look like.

Biography:

Paul Bellette is the Manager of the Data Processing team for Brisbane mining technology company Plotlogic, where he is responsible for the development of software that performs sensor fusion and interpretation of Hyperspectral scans to improve the efficiency of operations. He has previously worked for Brisbane based start-ups Roames (now part of Fugro) and Elara, working on a variety of remote sensing, modelling and optimisation applications. He has an undergraduate degree in Science from UTAS, with majors in Mathematics and Physics and a PhD at UQ in Engineering focusing on dynamics and control theory, where he also spent time as a postdoctoral researcher.