All Roads Lead to Rome? Exploring Representational Similarities Between Latent Spaces of Generative Image Models
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.
Cite
Text
Badrinath et al. "All Roads Lead to Rome? Exploring Representational Similarities Between Latent Spaces of Generative Image Models." ICML 2024 Workshops: SPIGM, 2024.Markdown
[Badrinath et al. "All Roads Lead to Rome? Exploring Representational Similarities Between Latent Spaces of Generative Image Models." ICML 2024 Workshops: SPIGM, 2024.](https://mlanthology.org/icmlw/2024/badrinath2024icmlw-all-b/)BibTeX
@inproceedings{badrinath2024icmlw-all-b,
title = {{All Roads Lead to Rome? Exploring Representational Similarities Between Latent Spaces of Generative Image Models}},
author = {Badrinath, Charumathi and Bhalla, Usha and Oesterling, Alex and Srinivas, Suraj and Lakkaraju, Himabindu},
booktitle = {ICML 2024 Workshops: SPIGM},
year = {2024},
url = {https://mlanthology.org/icmlw/2024/badrinath2024icmlw-all-b/}
}