Jeremy Rieussec
Computing science and operational research
Cohort 2021
Université de Montréal
Computing science and operational research
Cohort 2021
Université de Montréal
Computing science and operational research
Université de Montréal
Doctoral's
Sept. 2017 – Aug. 2019: Math Teacher in high school for classes equivalent to secondary 5, CEGEP and DEC in Paris, France. // 2015 – 2016: Preparation teaching exam in mathematics at University Paul Sabatier Toulouse III. Admitted at 2016 exam session, option probabilities and statistics. // 2014 – 2015: Research Master in Operational Research at ENAC (Ecole Nationale de l'Aviation Civile) in Toulouse, France. // 2012 – 2015: Engineer's degree in Computer science, programming, computer systems and aeronautics at ENAC (Ecole Nationale de l'Aviation Civile) in Toulouse, France.
PhD subject: study and development of optimization methods for the estimation of statistical models using maximum likelihood or least squares, in a context of big-data. I work on large-scale nonconvex stochastic optimization problems where the objective function is expressed as an expectation. Especially, the focus is given on problems such as maximum likelihood estimation and non-linear least squares regression. Stochastic variations of second-order methods where some approximation of the Hessian matrix is used within a trust-region framework are studied and tested. The methods proposed are tested on machine learning problems such as the training of multilayer perceptrons and convolutional networks on datasets like MNIST or CIFAR. Some more advanced tests on natural language processing with BERT models are also considered. Keywords: stochastic optimization, second-order, statistical learning, trust-region methods, Hessian-free, variable sample-path methods, Monte Carlo methods
An adaptive subsampled Hessian-free method for statistical learning in a trust-region framework
