What type of study is this?

This is a Quantitative Study study.

October 20, 2025Open Access

Define latent spaces by example: optimisation over the outputs of generative models

View Full Paper

Ask AI

Bookmark

Authors

This approach establishes surrogate latent spaces that optimise outputs in generative models, suggesting efficient task-specific control.

Our method allows for optimisation over outputs of generative models, enhancing control in applications.
Surrogate latent spaces are low-dimensional, Euclidean embeddings extracted from generative models with minimal computational cost.
This architecture-agnostic strategy generalises across various modalities, including images, audio, and structured objects.
Defining the axes via examples creates interpretable custom latent spaces tailored to specific downstream tasks.

Willis et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark