Disentangled Representations via Synergy Minimization

TL;DR

This paper proposes a method to learn disentangled data representations by explicitly minimizing informational synergy, which helps uncover true underlying factors more effectively than traditional independence-based approaches.

Contribution

It introduces a novel synergy minimization approach for disentangling factors of variation and provides a new benchmark task for evaluating such representations.

Findings

MinSyn representations successfully disentangle characters from word images.

Minimizing synergy outperforms independence-based methods in disentanglement.

New closed-form expressions for quantifying synergy are derived.

Abstract

Scientists often seek simplified representations of complex systems to facilitate prediction and understanding. If the factors comprising a representation allow us to make accurate predictions about our system, but obscuring any subset of the factors destroys our ability to make predictions, we say that the representation exhibits informational synergy. We argue that synergy is an undesirable feature in learned representations and that explicitly minimizing synergy can help disentangle the true factors of variation underlying data. We explore different ways of quantifying synergy, deriving new closed-form expressions in some cases, and then show how to modify learning to produce representations that are minimally synergistic. We introduce a benchmark task to disentangle separate characters from images of words. We demonstrate that Minimally Synergistic (MinSyn) representations correctly disentangle characters while methods relying on statistical independence fail.