The paper investigates the utility of conformal prediction sets in AI-advised decision-making. Through an experiment, it compares conformal prediction sets to displays of Top-1 and Top-k predictions for AI-advised image labeling. The study finds that the utility of these prediction sets for accuracy varies with the task difficulty. They result in accuracy on par with or less than Top-1 and Top-k displays for easy images but excel at assisting humans in labeling out-of-distribution images, especially when the set size is small. The paper also discusses the practical challenges of conformal prediction sets and their implications for real-world decision-making.
Publication date: 16 Jan 2024
Project Page: https://doi.org/XXXXXXX.XXXXXXX
Paper: https://arxiv.org/pdf/2401.08876
