Difference-in-Differences: Bridging Normalization and Disentanglement in PG-GAN

Abstract

What mechanisms causes GAN's entanglement? Although developing disentangled GAN has attracted sufficient attention, it is unclear how entanglement is originated by GAN transformation. We in this research propose a difference-in-difference (DID) counterfactual framework to design experiments for analyzing the entanglement mechanism in on of the Progressive-growing GAN (PG-GAN). Our experiment clarify the mechanisms how pixel normalization causes PG-GAN entanglement during a input-unit-ablation transformation. We discover that pixel normalization causes object entanglement by in-painting the area occupied by ablated objects. We also discover the unit-object relation determines whether and how pixel normalization causes objects entanglement. Our DID framework theoretically guarantees that the mechanisms that we discover is solid, explainable and comprehensively.

References in corpus (9)

• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• Self-Attention Generative Adversarial Networks
• Progressive Growing of GANs for Improved Quality, Stability, and Variation
• LSUN: Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loop
• Visualizing and Understanding Recurrent Networks
• Deep Convolutional Inverse Graphics Network
• GAN Dissection: Visualizing and Understanding Generative Adversarial Networks
• CXPlain: Causal Explanations for Model Interpretation under Uncertainty
• Counterfactuals uncover the modular structure of deep generative models