Machine Learning

In neuroscience, the choice of the machine learning algorithm and its hyperparameters, are often arbitrary and frequently do not correspond to the statistical properties of the data being analysed. In this project, I used an autoML approach to explore a vast space of models and their hyperparameters, and in an entirely data-driven fashion, find the pipeline that most accurately predicts brain age.

Different algorithms have been used to predict brain age from structural images. However, every study uses different algorithms on different datasets, making it difficult to compare the results among each other. This project aimed to bridge this gap and rigorously compare the most commonly used algorithms to predict brain age and to compare the impact of using different inputs on the precision of the analysis.

Although different computational models have been developed to investigate the neural dynamics of the brain, less is known about the brain-environment interaction and how the latter influence brain dynamics. Therefore, we developed a virtual agent, whose activity is driven by a computational model, and analysed how the environment influenced its neural dynamics.

This project aimed to develop and test computational models of learning that quantify how people learn under social uncertainty .We used a combination of methods, which included: hierarchical Bayesian model of learning, model-based fMRI and DCMs