MEET THE LECTURER: Dr Davide Ferrari
The University of Melbourne
Dr Ferrari earned his PhD in Statistics at the University of Minnesota, US. He is an academic statistician and currently has a continuing position as a Senior Lecturer in statistics at the School of Mathematics and Statistics, University of Melbourne. He is also an Associate Research Fellow at the ARC Centre of Excellence for Mathematical and Statistical Frontiers. Davide´s main research interests are minimum divergence estimation methods, robust statistics and model selection. Recently, he has started working on model selection and estimation methods for intractable likelihoods in the context of complex biological data which allows him to combine his interest in likelihood-based methods with his interest in biomedical applications.
Dr Davide Ferrari will be lecturing at AMSI Winter School 2017, delivering a course on “Model Selection and Inference for High-Dimensional Data”.
1. Can you tell us about your work? What drives your interest in this field?
Statistics helps us understand how information might have been generated from real-life processes, provides us with powerful tools to extract such information from observed data and ultimately enables us to learn about such processes. My research is ultimately driven by the desire to discover what information really is from a mathematical viewpoint. This is a continuous discovery process leading to new statistical methods which in turn foster progress in various areas of empirical research.
2. What are the most interesting “big questions” or challenges facing researchers in your area?
Modern technologies allow scientists to collect data of unprecedented size. For high dimensional data, however, the question “Which statistical model is the best?” is not answered unequivocally by existing methods. One very important challenge in this context which not fully addressed by current methods is how to describe and control the model selection uncertainty. Simply put, why we do not have yet confidence intervals in the context of model selection? These days no serious data analyst would think to report estimates without confidence intervals or some other appropriate error measure.
However, more often than not scientist and practitioners alike completely ignore the error related to model selection, which may lead to unreliable models, especially in the context of complex data. Another important issue in modern statistics is that traditional methods such as maximum likelihood estimation are severely limited since the models in big data applications are often computationally intractable or impossible to specify. Thus, there are a number of urgent questions related to the need to develop new classes of inference methods to computationally feasible statistical inferences/predictions for complex data problems with a statistical performance comparable to that of maximum likelihood.
3. What are some key industry applications of your work?
My industry applications include prediction of tropical cyclones (with the Bureau of Methereology, AUS), estimation of crime rate in residential buildings (with Epsilon Security, AUS), design and analysis of clinical trials for vaccine deveolopment (with Synthiron, US). I often apply my model selection methods and inference methods for intractable likelihoods in cancer research.
4. What do you consider your biggest achievement to date?
The development of a theoretical understanding of non-additive information measures (q-entropies) in the context of inference and their role in statistics.
5. Why did you become a mathematician/statistician?
First, I want to understand what information is from a mathematical viewpoint and discover new and better ways to characterise it. Second, I am driven by the desire to apply such discoveries to practically useful problems. Some of my findings can hopefully boost the rate of progress in many areas of empirical research leading to increased well-being of the community at large (e.g. new cancer treatments).
6. Do you have any advice for future researchers?
A statistician, like an artist, has to be passionate and devoted. Without strong drive and internal motivation you will get stuck, but if you genuinely love statistics and mathematics you will obtain the greatest satisfaction and reward from solving challenging problems. Never stop to dare and ask new questions: the more difficult the questions are, the higher is the final reward.