Max Hinne

As of August 2019, I am an assistant professor at the Department of Artificial Intelligence, at the Radboud University, Nijmegen, The Netherlands.

There are few objects more interesting than the brain. It has just the right structure and dynamics that results in large-scale information processing, from which our complex cognition emerges and using which it ultimately studies itself.

In my research, I use Bayesian modelling & statistics to learn latent (network) structures in the human brain, via neuroimaging modalities such as fMRI and dwMRI, and to use these as predictive models. I am particularly interested in Bayesian nonparametric models (such as Gaussian processes), in which the complexity of the model is learned from data instead of provided a priori. Other, related, interests incude (quasi-)causal inference, predictive processing, and dynamical systems.

My contact information is listed below.

“All stable processes we shall predict. All unstable processes we shall control.”

John von Neumann

Master thesis projects

Interested in doing a master thesis on topics such as applying or developing new Bayesian nonparametric methods to neuroimaging data? Send me an email!


PhD thesis

On June 9th 2017, I defended my doctoral thesis entitled Bayesian Connectomics: a probabilistic perspective on brain networks with a cum laude distinction.




The Bayesian Connectomics Toolbox contains Matlab/MEX scripts that implement MCMC approximations for:

Basic usage of each of the scripts is demonstrated in demo.m. Feel free to email me with any questions or comments.

Latent space modeling

Matlab code and Stan models are available for the latent space model that was used for link prediction. Running the sample requires that you have Stan and the MatlabStan interface installed.


Code for the causal inference / effective connectivity estimation using GP CaKe is available here. There is a tutorial for GP CaKe in the form of a series of blog post, which you can read on the MindCodec website.

Data sets

Two data sets are currently available:

  1. Probabilistic streamline counts based on diffusion-weighted MRI, for 162 regions (FreeSurfer parcellation: 74 cortical regions per hemisphere, 7 subcortical regions per hemisphere). These data were used in the paper regarding the clustered connectome.
  2. Functional MRI BOLD response timeseries, as well as probabilistic streamline counts based on diffusion-weighted MRI, for 14 subcortical regions (7 per hemisphere). These data were used in the paper on conditional independence and data fusion.
Credit for collecting these data goes to Erik van Oort.

Contact information


Google Scholar
Max Hinne





Visiting Address AI

Room B.00.67B
Montessorilaan 3
6525 HR Nijmegen

The Bayesian party parrot approves.