Differentiable Set Partitioning

Thomas M. Sutter, Alain Ryser, Joram Liebeskind, Julia E Vogt

Keywords: random partition model, continuous relaxation, reparameterization, generative models, vae, representation learning, weak supervision, variational clustering, deep learning

TL;DR: We propose a differentiable random partition model to be integrated into modern gradient-based optimization pipelines.

Abstract:

Partitioning a set of elements into an unknown number of mutually exclusive subsets is essential in many machine learning problems.
However, assigning elements, such as samples in a dataset or neurons in a network layer, to an unknown and discrete number of subsets is inherently non-differentiable, prohibiting end-to-end gradient-based optimization of parameters.
We overcome this limitation by proposing a novel two-step method for inferring partitions, which allows its usage in variational inference tasks.
This new approach enables reparameterized gradients with respect to the parameters of the new random partition model.
Our method works by inferring the number of elements per subset and, second, by filling these subsets in a learned order.
We highlight the versatility of our general-purpose approach on two different challenging experiments: multitask learning and inference of shared and independent generative factors under weak supervision.

Submission Number: 19