Why do neural networks make the decisions they do? This workshop offers a hands-on introduction to latent spaces—the hidden layers where networks encode meaning — and shows how understanding them allows us to steer models toward desired outcomes. Steering involves finding directions in latent space that correspond to meaningful changes in output, such as modifying facial attributes in images or shifting tone and sentiment in text. These techniques offer powerful tools for model control, interpretability, and customization. 

Participants will gain practical, hands-on experience with real-world examples across domains, helping them understand how to influence model behavior and make AI systems more transparent and aligned with human intent. Whether you’re focused on generative models, bias mitigation, or creative applications, this workshop offers valuable skills for building smarter, more controllable AI.

Program

Introduction, background, and motivation (30 minutes)

Hands-on exercises in groups (50 minutes)
participants need laptop with access to internet, no additional setup required

Discussion + wrap-up (10 minutes)

Level

Introductory: suitable for beginners with little to no prior knowledge of the subject.

Organizers

• Lenka Tětková, Postdoc, DTU Compute
• Anders Gjølbye Madsen, PhD student, DTU Compute
• Akhil Arora, Assistant professor, Aarhus University and SODAS
• Sarah Masud, Postdoc, DIKU (KU)
• Lars Kai Hansen, Professor, DTU Compute