Bayesian Active-Learning: Time-Efficient Testing in Clinical Practice and Research Seminar

Time-efficiency is an important factor for every test. Bayesian active learning provides a framework for achieving this goal by using probabilistic models that estimate the outcome during the test, and choosing parameters for further trials that are informative. These tests lead to more knowledge or higher accuracy in the same time. In the healthcare sector with often strict time constraints per patient, a faster method that does not compromise on accuracy can make a test applicable at all.

This talk gives a short introduction to Bayesian active learning, in particular using Gaussian Processes (GPs) as probabilistic models. Furthermore, several hearing tests are presented as example applications: The audiogram and an auditory-filter test (GP classification), equal-loudness-contour test (GP regression), and a dead-region test (based on a hearing model and not using a GP). All these tests were evaluated using human subjects, and they outperformed conventional methods.