Covering up bias in CelebA-like datasets with Markov blankets: A post-hoc cure for attribute prior avoidance

Abstract

Attribute prior avoidance entails subconscious or willful non-modeling of (meta)attributes that datasets are oft born with, such as the 40 semantic facial attributes associated with the CelebA and CelebA-HQ datasets. The consequences of this infirmity, we discover, are especially stark in state-of-the-art deep generative models learned on these datasets that just model the pixel-space measurements, resulting in an inter-attribute bias-laden latent space. This viscerally manifests itself when we perform face manipulation experiments based on latent vector interpolations. In this paper, we address this and propose a post-hoc solution that utilizes an Ising attribute prior learned in the attribute space and showcase its efficacy via qualitative experiments.

arXiv commentAccepted for presentation at the first workshop on Invertible Neural Networks and Normalizing Flows (ICML 2019), Long Beach, CA, USA

References in corpus (7)

• Progressive Growing of GANs for Improved Quality, Stability, and Variation
• Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
• NICE: Non-linear Independent Components Estimation
• Fairness in Machine Learning: Lessons from Political Philosophy
• DNA-GAN: Learning Disentangled Representations from Multi-Attribute Images
• GeneGAN: Learning Object Transfiguration and Attribute Subspace from Unpaired Data
• Generative adversarial interpolative autoencoding: adversarial training on latent space interpolations encourage convex latent distributions