Human in the Loop Bayesian Optimization for ALL

Co located with CHI 2026, Barcelona

Human in the Loop Bayesian Optimization supports sample efficient search in complex HCI design spaces by integrating user feedback into model driven decision making. This 90 minute workshop combines core concepts, a live Python and Unity demonstration, and guided activities to help participants frame their own optimization workflows. The focus is on balancing subjective outcomes such as trust and workload with objective performance, while managing noise, multiple objectives, and fairness across user groups.

Organizers

Pascal Jansen, Institute of Media Informatics, Ulm University, Germany, pascal.jansen@uni-ulm.de
Mark Colley, UCL Interaction Centre, University College London, United Kingdom, m.colley@ucl.ac.uk

Submission

Short position papers, 2 to 4 pages, ACM single column format. Describe a design problem for optimization, lessons from data driven or adaptive design, or a critical perspective on optimization in HCI. Accepted papers will be published on the workshop website. See Participate for details.

Intended audience

Researchers, PhD students, UX and industry practitioners who want to integrate optimization into design and evaluation. No prior BO experience required, basic Python or C# helps.

Prerequisites

Laptop with Python 3.13 or later, BoTorch v0.15.1 or later, optional Unity 6.2 for the live link, familiarity with Jupyter recommended.

Schedule, 90 minutes

| Phase | Duration | | — | — | | Phase I, Introduction and motivation | 10 min | | Phase II, Live demo, foundations of BO | 20 min | | Phase III, Small group use case mapping | 15 min | | Break | 5 min | | Phase IV, Panel discussion, MOBO | 20 min | | Phase V, Pipeline sketching and short reports | 15 min | | Phase VI, Wrap up and next steps | 5 min |

Accessibility

Request real time captioning through CHI services
Distribute tagged PDF slides with alt text and high contrast
Provide short captioned demo videos
Use low barrier templates and shared cloud notebooks

Materials

GitHub with annotated Jupyter notebooks, BoTorch examples, Unity link, datasets, evaluation templates
Slide decks, quick start guides, troubleshooting notes
Curated bibliography
Post workshop summary report

Call for participation

Designing interactive systems requires balancing competing objectives such as speed and accuracy or efficiency and usability. Human in the Loop BO provides a principled, sample efficient path to explore trade offs with fewer trials. This session provides the foundations, a live pipeline, and guided tasks so you leave with an actionable plan for your project.

Attendance

Target, 15 to 25 in person participants.