How meaningful are similarities in deep trajectory representations?

Saeed Taghizadeh, Ábel Elekes, Martin Schäler, Klemens Böhm

Abstract: Highlights•We address the meaningfulness of similarity values in deep trajectory models.•We evaluate the robustness of the deep trajectory model to parameterization.•Similarity values are different for models trained with different parameters.•The deep model captures different semantics of similarity than the classical models.•T2vec is faster, and it is better for clustering than classical similarity models.