All Roads Lead to Rome? Exploring Representational Similarities Between Latent Spaces of Generative Image Models

2024/07/18 by Badrinath, Charumathi, Bhalla, Usha, Oesterling, Alex +2
#FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2407.13449

Abstract

Do different generative image models secretly learn similar underlying representations? We investigate this by measuring the latent space similarity of four different models: VAEs, GANs, Normalizing Flows (NFs), and Diffusion Models (DMs). Our methodology involves training linear maps between frozen latent spaces to "stitch" arbitrary pairs of encoders and decoders and measuring output-based and probe-based metrics on the resulting "stitched'' models. Our main findings are that linear maps between latent spaces of performant models preserve most visual information even when latent sizes differ; for CelebA models, gender is the most similarly represented probe-able attribute. Finally we show on an NF that latent space representations converge early in training.

Related

• Comparing the latent space of generative models
• Generative Models: What Do They Know? Do They Know Things? Let's Find Out!
• LASERS: LAtent Space Encoding for Representations with Sparsity for Generative Modeling
• On the Transformation of Latent Space in Autoencoders
• Complexity Matters: Rethinking the Latent Space for Generative Modeling