von Mises-Fisher Sampling of GloVe Vectors

Walid Bendada, Guillaume Salha-Galvan, Romain Hennequin, Théo Bontempelli, Thomas Bouabça, Tristan Cazenave

Keywords: von Mises-Fisher sampling, Boltzmann sampling, exploration, large action sets, scalability, hyperspherical embedding vectors

TL;DR: We present the first experimental validation of the von Mises Fisher Exploration (vMF-exp) sampling method's key theoretical and scalability properties on a large-scale and publicly available real-world dataset.

Abstract: A recent publication introduced von Mises-Fisher exploration (vMF-exp), a scalable sampling method for exploring large action sets in reinforcement learning problems where hyperspherical embedding vectors represent these actions. We present the first experimental validation of vMF-exp’s key theoretical and scalability properties on a publicly available real-world dataset, confirming the potential of this method.

Submission Number: 3