Learning to Be Guided: Shaping Latent Geometry for Classifier-Free Guidance
Classifier-free guidance (CFG) is a standard component of conditional diffusion models, and guided generation performance has become the de facto measure of state-of-the-art models.
ProofPaper ↗
Key points
- Recent work on latent diffusion models seeks to learn latent spaces that are easier to model with diffusion.
- We study this arrangement under discrete class conditions, termed class geometry, and investigate what class geometry allows a latent space to work well with CFG.
- Under a Gaussian class model, we show that shifting variance from intra- to inter-class at fixed total variance makes classes more distinguishable under noise, and derive a noisy posterior objective whose per-sample descent direction is proportional to the oracle CFG correction.
- Based on the analysis, we propose Prototype Regularization (PR), a lightweight surrogate for the intractable noisy posterior objective, which applies a cosine prototype classifier to spatially pooled latents.
Sources (1)
- [1]Learning to Be Guided: Shaping Latent Geometry for Classifier-Free GuidancearXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 03:39 AM
Classifier-free guidance (CFG) is a standard component of conditional diffusion models, and guided generation performance has become the de facto measure of state-of-the-art models.
Recent work on latent diffusion models seeks to learn latent spaces that are easier to model with diffusion.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 16, 2026MOSS-Image Technical Report
- Oct 16, 2026Persistent Planar Memory for Video World Models
- Oct 15, 2026Reinforcement Learning-driven Diffusion Models: Language, Vision, and Beyond
- Oct 15, 2026SUPRA-TEX: 3D Generation with Super-Aligned Texture
- Oct 14, 2026Purify Before You Align: Improving Representation Alignment for Diffusion Models
- Oct 14, 2026Modeling and Understanding Neural Weights of Humanoid Policies with Diffusion