Vincent Spruyt: Loc2Vec: Self-supervised metric learning through triplet-loss

Self-supervised learning is an increasingly popular technique to learn meaningful representations of data when no labels are available. A related problem is that of learning a mapping from raw input data into a metric space, where distances between latent data points are proportional to the semantic similarity between the original data instances. In this talk, we show how triplet-loss can be used to train a neural network in a self-supervised manner by applying it to location data. The result is a transform