Eindhoven University of Technology
EDL-P16-25 P5: Deep Learning for Human and Animal Health (HAAH)
Personalization of Hearing Aids
Hearing aid (HA) users prefer different audio processing algorithms, depending on the acoustic context and their personalized hearing loss profile. Currently, the approach to tuning HA parameters relies on the user visiting an audiologist who manually tunes the HA in her clinic – a setting far removed from the lived experience of the HA user. My research aims at developing an intelligent agent that can learn optimal parameter settings in situ from user appraisals to circumvent this problem.
Given the target demographic, this requires learning from a very small amount of sparse feedback signals. To that end we draw inspiration from one of the most data efficient systems we know: The human brain. Active Inference is a framework the describes the kinds of computations the brain might perform when engaging in learning and decision making. In my work, I apply Active Inference to the problem of designing intelligent agents with the aim of developing an agent that can meet the strict data efficiency requirements of in situ HA tuning.