Exploring Dimensionality Reduction on Semi-structured Photos — A Closer Look at Exactitudes
Sam D. Verkoelen
Deep learning or dimensionality reduction has shown unprecedented results in contemporary researches. However, these systems are often referred to as black boxes. This exploratory research will look at dimensionality reduction via Restricted Boltzmann Machines in a set of small experiments. These experiments were performed on the semi-structured photos of the "exactitudes" project, resulting in novel insights and visualisations on how to extract meaningful information from a deep networks. These insights do not only help to understand what such systems learns but it also provides a glimpse of the underlaying structure of the dataset.
Sam D. Verkoelen, "Exploring Dimensionality Reduction on Semi-structured Photos — A Closer Look at Exactitudes", Master's Thesis for the Media Technology programme, Leiden University (The Netherlands), 2015