The lack of individualized fitting of hearing aids results in many patients never getting the intended benefits, in turn causing the devices to be left unused in a drawer. However, living with an untreated hearing loss has been found to be one of the leading lifestyle related causes of dementia and cognitive decline. Taking a radically different approach to personalize the fitting process of hearing aids, by learning contextual preferences from user-generated data, we in this paper outline the results obtained through a 9-month pilot study. Empowering the user to select between several settings using Internet of things (IoT) connected hearing aids allows for modeling individual preferences and thereby identifying distinct coping strategies. These behavioral patterns indicate that users prefer to switch between highly contrasting aspects of omnidirectionality and noise reduction dependent on the context, rather than relying on the medium “one size fits all” program frequently provided by default in hearing health care. We argue that an IoT approach facilitated by the usage of smartphones may constitute a paradigm shift, enabling continuous personalization of settings dependent on the changing context. Furthermore, making the user an active part of the fitting solution based on self-tracking may increase engagement and awareness and thus improve the quality of life for hearing impaired users.
Deep neural networks (DNN) and Gaussian processes (GP) are two powerful models with several theoretical connections relating them, but the relationship between their training methods is not well understood. In this paper, we show that certain Gaussian posterior approximations for Bayesian DNNs are equivalent to GP posteriors. This enables us to relate solutions and iterations of a deep-learning algorithm to GP inference. As a result, we can obtain a GP kernel and a nonlinear feature map while training a DNN. Surprisingly, the resulting kernel is the neural tangent kernel. We show kernels obtained on real datasets and demonstrate the use of the GP marginal likelihood to tune hyperparameters of DNNs. Our work aims to facilitate further research on combining DNNs and GPs in practical settings.