Program

Please note that the program is tentative and subject to change anytime (even during the workshop).

We are very pleased to have two fine keynote speakers for the workshop: Adam Kowalczyk of GeBiS, Diversity Arrays Technology, and the University of Melbourne Centre for Neural Engineering, and Wei Bian of the UTS Centre for Quantum Computation and Intelligent Systems.

Abstracts for their talks are below:

Wei Bian: Asymptotic generalisation analysis of linear models with inverse covariance: a random matrix theory based perspective.

Abstract: Linear models such as Least Square Regression (LSR) and Linear Discriminant Analysis (LDA) are important statistical models for data analysis. However, these models do not generalise well in large-dimensional cases, where the dimensionality p of data is proportional to the training sample size n. A major reason is the inconsistent estimation of the inverse covariance matrix. In this talk, we introduce basic tools from the random matrix theory, especially the results on the spectra of sample covariance, to analyse the generalisation of linear models with inverse covariance. Thanks to the nature of random matrix theory, the obtained asymptotic generalisation bounds of LSR and LDA are considerably tight with high confidence, as verified by numerical simulations. Furthermore, the results motivate a block-diagonal regularisation for LSR and LDA in the high-dimensional settings.

Adam Kowalczyk: Genome-Wide Discovery of Disease Gene Mutations, their Interactions, and their Biomarker Proxies - a challenge of predictive learning when  n,p→∞  & n/p→0

Abstract: The personalised medicine and precision agriculture pose a number of non-standard big-data challenges.  In particular, they demand practical solutions of predictive modelling tasks with both the number of samples n and the dimensionality of variables p roughly diverging to ∞ but their ratio n/p converging to 0. The development of efficient data-mining techniques for extraction of knowledge and predictive modelling for such studies constitutes currently the major bottle-neck in realisation of hopes and promises of the expected genomic revolution.  We discuss some of our research results, concentrating on computational, statistical and machine learning challenges.