Latent Space: The AI Engineer Podcast cover image

ICLR 2024 — Best Papers & Talks (ImageGen, Vision, Transformers, State Space Models) ft. Durk Kingma, Christian Szegedy, Ilya Sutskever

Latent Space: The AI Engineer Podcast

NOTE

Decomposing Concepts and Images with Diffusion Models

The work aims to interpret the internal representations of concepts generated by diffusion models, such as diverse images created by a concept painter. The main objectives are to decompose concepts into features and to break down individual images into the specific features used for their generation. To achieve this, the study will use the stable diffusion vocabulary, a rich and semantic token set based on the clip vocabulary. By training a lean MLP to map tokens to coefficients, the research intends to analyze the internal features utilized by diffusion models in representing concepts and generating images.

00:00
Transcript
Play full episode

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
App store bannerPlay store banner