While the national Mastermath program allows you to follow courses all across the country, the master thesis project should make the difference in your choice of university. Have a look at a list of exciting thesis topics that you could explore at the Vrije Universiteit Amsterdam.

Project: Asymptotics of nonparametric regression using sparse neural networks with ReLU activation
Supervisor: Eduard Belitser
Short description: A paper by Schmidt-Hieber (2020) was studied in great detail and the results of that paper were improved in a couple of aspects. The most important contribution by the student was the generalization of the original model to a setting with Markov chain regressors and sub-gaussian errors.

Project: Model reduction in probabilistic forecasting
Supervisor: Svetlana Dubinkina
Short description: Standard techniques in model reduction of dynamical systems involve decomposition of the phase space into stable and unstable Lyapunov directions, and are based on a single trajectory. On the other hand, in probabilistic forecasting, it is an ensemble of trajectories that is forecasted over time. In this project, the student develops model reduction for ensemble forecasting. 

Project: A graph-theoretic framework for neural field models on the human connectome
Supervisor: Rikkert Hindriks
Short description: In this thesis a framework is proposed to define neural field models on graphs. The spatiotemporal power spectrum is calculated on the human connectome and compared with resting-state BOLD-fMRI data.

Project: Power with partially exact matching' [causal inference]
Supervisor: Stephanie van der Pas
Short description: Propensity score matching is a hugely popular causal inference method. In this project we investigate how power is affected by the number of variables on which exact matching is required, both in low- and high-dimensional settings.

Project: Stability analysis of Markov Chains
Supervisor: Wouter Kager
Short description: Markov processes have certain so-called stability criteria, under which a process is (positive) recurrent. In this project we investigate the measure-theoretical foundations of these criteria, which lead to a unified treatment, in particular in situations without an invariant distribution 

Project: A critical look at e-values
Supervisor: Ronald Meester
Short description: In this project we investigate  the pros and contra’s of e-values, which are supposed to be an alternative to the much criticized p-values. Many new examples are given and studied. 

Project: Filtering and parameter estimation in the Jansen-Rit model
Supervisor: Frank van der Meulen
Short description: The Jansen-Rit model is a neural population model of a local cortical circuit. In this project various ways to estimate unknown  parameters are investigated, such as the unscented Kalman filter. 

Project: Revisiting Vine Copula Imputation
Supervisor: Paulo Andrade Serra
Short description:  In this project the student studied whether Vine Copula imputation procedures provide enough of a gain in accuracy relative to time spent compared to simpler methods. Vine Copula imputation was compared to Gaussian Copula imputation, and benchmarked these against other popular methods.

Project: Nonparametric mean-covariance estimation in a periodic setting
Supervisor: Paulo Andrade Serra
Short description: A nonparametric regression model with short-range dependent correlated errors was investigated in a periodic setting. Estimators for the mean function and for the covariance structure were proposed and their consistency and convergence rate was investigated. It was shown that the mean estimator performs optimally in the short-range noise setting, even if the covariance structure is