UI for Exploring Metrics for Generative Models' Latent Spaces

Problem Statement

The increased use of generative models has elevated human-AI collaboration to new heights. Although there has been extensive research on the evaluation metrics of generative models, these have not yet been leveraged for human exploration of the latent space of such models. Exploring latent spaces remains a significant challenge, particularly in terms of how users interact with these spaces through different user interfaces. This work investigates the influence of user interfaces on users’ exploration of generative model latent spaces, aiming to develop more intuitive and effective exploration tools. Specifically, in this thesis, we aim to build a visualization tool for understanding the metrics related to GANs' latent space—capturing how users explore, navigate, and engage with generative outputs. By focusing on UI design and interaction, this work sheds light on user-centric exploration patterns and provides practical insights for designing accessible and engaging interfaces. The findings contribute to a deeper understanding of user interaction with generative models and facilitate their broader adoption and application across various domains.

Tasks

• Perform a literature review. Identify key challenges and gaps in current approaches to GAN's latent space Exploration
• Build a visulaization tool from the python package provided for different metrics
• Conduct a user studys
• Do statistical evaluations
• Write a thesis and present your findings in the Disputationsseminar
• (Optional:) co-write a research paper