Causally Disentangled Generative Variational AutoEncoder

TL;DR

This paper introduces Causally Disentangled Generation (CDG), a supervised VAE-based approach that learns causally disentangled representations and generates outcomes accordingly, supported by a new evaluation metric and empirical validation.

Contribution

It proposes a novel supervised VAE method for causal disentanglement, identifies key conditions for achieving CDG, and introduces a universal metric for evaluation.

Findings

Supervised regularization alone is insufficient for CDG.

Necessary and sufficient conditions for CDG are explored.

Empirical results validate the approach on image and tabular data.

Abstract

We present a new supervised learning technique for the Variational AutoEncoder (VAE) that allows it to learn a causally disentangled representation and generate causally disentangled outcomes simultaneously. We call this approach Causally Disentangled Generation (CDG). CDG is a generative model that accurately decodes an output based on a causally disentangled representation. Our research demonstrates that adding supervised regularization to the encoder alone is insufficient for achieving a generative model with CDG, even for a simple task. Therefore, we explore the necessary and sufficient conditions for achieving CDG within a specific model. Additionally, we introduce a universal metric for evaluating the causal disentanglement of a generative model. Empirical results from both image and tabular datasets support our findings.