Hi! I’m a cognitive neuroscientist working as a postdoc at the Werner Reichardt Centre for Integrative Neuroscience and the Max Planck Institute for Intelligent Systems in Tübingen. The aim of my research is to understand how our brains enable us to perceive and make correct inferences about our surroundings given the very limited information that is available to our senses.
I use modern neuroimaging techniques and psychophysical modeling to understand the complex relationships between brain, mind, and the outside world. To address my research questions, I have been using color vision as a model system as well as motion perception.
More recently, I became interested in computer vision. The fascinating progress in artificial intelligence raises the question how comparisons between biological and artificial cognitive systems (i.e., brains and machines) can enhance our understanding of the human mind and behavior.
I recently wanted to fit a GLM with a large design matrix to a high-dimensional dataset with few samples, as is typical in fMRI. Since the design matrix had a lot more predictors than observations, I naturally used a regularized regression model with integrated hyperparameter tuning, i.e. scikit-learn’s RidgeCV. Instead of tediously fitting a model and optimizing its parameters for each time series one after the other, I wanted to take advantage of the fact that all scikit-learn classifiers support multitarget classification/regression by default. However, I noticed that RidgeCV finds the optimal hyperparameter across all the...
Drop me an email if you would like to know more.