Forced to Learn: Discovering Disentangled Representations Without Exhaustive Labels

Alexey Romanov, Anna Rumshisky

Abstract: Learning a better representation with neural networks is a challenging problem, which was tackled extensively from different prospectives in the past few years. In this work, we focus on learning a representation that could be used for clustering and introduce a novel loss component that substantially improves the quality of produced clusters, is simple to apply to an arbitrary cost function, and does not require a complicated training procedure.

TL;DR: A novel loss component that leads to substantial improvement of KMeans clustering over the learned representations.

Conflicts: cs.uml.edu