aiqbench analysis
This research introduces ConDA, a plug-and-play geometry layer that applies contrastive learning to diffusion model latents using auxiliary variables like time and stimulation parameters. By aligning latent directions with underlying dynamical factors, the method enables smooth interpolation, extrapolation, and counterfactual editing across diverse domains such as fluid dynamics and neural imaging without leaving the original diffusion space. The approach demonstrates superior interpretability and control compared to linear traversals or standard conditioning baselines, highlighting how explicit contrastive geometry can unlock structured latent representations in generative models.
Contrastive Diffusion Alignment: Learning Structured Latents for Controllable Generation
arXiv:2510.14190v3 Announce Type: replace Abstract: Diffusion models excel at generation, but their latent spaces are high dimensional and not explicitly organized for interpretation or control. We introduce ConDA (Contrastive Diffusion Alignment), a plug-and-play geometry layer that applies contrastive learning to pretrained diffusion latents using auxiliary variables (e.g., time, stimulation parameters, facial action units). ConDA learns a low-dimensional embedding whose directions align with underlying dynamical factors, consistent with recent contrastive learning results on structured and disentangled representations. In this embedding, simple nonlinear trajectories support smooth interpolation, extrapolation, and counterfactual editing while rendering remains in the original diffusion space. ConDA separates editing and rendering by lifting embedding trajectories back to diffusion latents with a neighborhood-preserving kNN decoder and is robust across inversion solvers. Across fluid dynamics, neural calcium imaging, therapeutic neurostimulation, facial expression dynamics, and monkey motor cortex activity, ConDA yields more interpretable and controllable latent structure than linear traversals and conditioning-based baselines, indicating that diffusion latents encode dynamics-relevant structure that can be exploited by an explicit contrastive geometry layer.
Authors:Ruchi Sandilya, Sumaira Perez, Charles Lynch, Lindsay Victoria, Benjamin Zebley, Derrick Matthew Buchanan, Mahendra T. Bhati, Nolan Williams, Timothy J. Spellman, Faith M. Gunning, Conor Liston, Logan Grosenick
Abstract:Diffusion models excel at generation, but their latent spaces are high dimensional and not explicitly organized for interpretation or control. We introduce ConDA (Contrastive Diffusion Alignment), a plug-and-play geometry layer that applies contrastive learning to pretrained diffusion latents using auxiliary variables (e.g., time, stimulation parameters, facial action units). ConDA learns a low-dimensional embedding whose directions align with underlying dynamical factors, consistent with recent contrastive learning results on structured and disentangled representations. In this embedding, simple nonlinear trajectories support smooth interpolation, extrapolation, and counterfactual editing while rendering remains in the original diffusion space. ConDA separates editing and rendering by lifting embedding trajectories back to diffusion latents with a neighborhood-preserving kNN decoder and is robust across inversion solvers. Across fluid dynamics, neural calcium imaging, therapeutic neurostimulation, facial expression dynamics, and monkey motor cortex activity, ConDA yields more interpretable and controllable latent structure than linear traversals and conditioning-based baselines, indicating that diffusion latents encode dynamics-relevant structure that can be exploited by an explicit contrastive geometry layer.
Machine Learning (cs.LG)
arXiv:2510.14190 [cs.LG]
(or arXiv:2510.14190v3 [cs.LG] for this version)
Full text under the publisher's open licence · 202 words
Related coverage
Identifiable World Models from Pretrained Diffusion Representations
arXiv · Oct 7, 2026
From Alignment to Synthesis: Contrastive Volumetric Grounding for Text-to-CT Generation
arXiv · Sep 17, 2026
Evaluating task vectors, unlearning and inoculation
LessWrong · Sep 20, 2026
[AINews] Here are 6 Clones of Jev in 2 days
Latent Space · Sep 19, 2026