An optimizer adjusts the colors of a photo and asks what you think. How it asks is the subject of the paper: the same person tells the optimizer different things through a rating scale, a choice between two, or a ranking. Pick an interface, give feedback for a few rounds, and compare.
Your runs
About this demo
The six interfaces and the photo color enhancement task are the ones used in the study (N=360) reported in the paper. The demo runs entirely in your browser: nothing you do here is recorded or sent anywhere.
To run without a server, the optimizer is a compact reimplementation of the study's Bayesian optimizers rather than the original BoTorch code. It converts your feedback the same way: ratings train a utility model, choices and rankings become pairwise comparisons, “I don't know” answers are left out, and a tie in a ranking counts in both directions. The study's source code, data, and analysis are in the repository.
Photos by Yuki Koyama, released under CC0 with Sequential Gallery.