Louis-Alexandre Leger

Gaussian Mixture Models are universal in their ability to approximate continuous densities. We show the inferred state connectomes of the spinal cord with the unsupervised Graph Laplacian Mixture Model. We find laplacians highly correlated with prior knowledge despite a poor signal and show the limits of spinal cord fMRI by parametrizing the feasibility of unsupervised clustering with multivariate Gaussians.

Comparison of Vision Transformers versus Convolutional Neural Networks, modulating network complexity proportional to the number of learnable parameters, attention weights visualization, "patch-mer" image tokenization.

A data analysis investigation into the causality of actors influence on a movie's success. Using the IMDB dataset, we evaluated the effect and analysed confounders such as gender, movie genre, movie directors and other actors.

Created a homemade VR system with blob detection, stereo distance calculation and google carboard device in order to play an augmented reality version of Pictionary. Players would draw on a canvas and other players would guess