What if we could use Gaussian processes to automatically describe time series in a human understandable way ?
(Multivariate) Gaussian processes are distributions over an ensemble of functions. They are entirely described by their mean vector :
and theirs covariance functions :
The use of deep neural networks has increased significantly in recent years. It is probably due to the improvement of cpu and gpu’s calculation abilities which allows DNNs to achieve state-of-the-art results in most areas, especially in the field of image classification. Unfortunately, the opacity of DNNs makes them difficult to interpret, hence they are not widely adopted in critical fields such as military or medical ones. This last few years, lot of research have focus on trying to understand how DNNs works and how to make them less obscure in order to increase users trust. This new field, called…
Research Intern @Eura Nova. Working on CNN calibration and interpretability