I'm a former academic (ex-Senior Lecturer at the University of Waikato) with a PhD in Computer Science/Probabilistic Machine Learning from the University of Birmingham (UK).
I am especially interested in high-dimensional small sample size problems $(n \ll p)$ which arise in many practical contexts: What properties of multivariate data make a problem "easy" in the sense that - given such structure - we can still build good models using only a small random sample of such data? 'Sparsity' is the best-known example of such intrinsic low-dimensionality, but for linear classifiers it turns out there are various other lucky structural properties that effectively make a problem easier than it looks: https://www.jair.org/index.php/jair/article/view/11506