Disentangling peripheral and central contributions to impaired perception via inversion of the neural code
Project ID: 2228cd1439 (You will need this ID for your application)
Research Theme: Information and Communication Technologies
UCL Lead department: Ear Institute
Lead Supervisor: Nicholas Lesica
Project Summary:
Why this research is important
Hearing loss is one of the world’s most widespread and disabling chronic conditions. Its complexity poses a fundamental problem, with damage to multiple cochlear structures that in turn elicits plastic changes in multiple central auditory brain structures. Fortunately, we can now engage with hearing loss in its full complexity: deep learning can be used to characterize the transformation from sound to neural activity patterns and vice-versa without any simplifying assumptions. By studying the inverse problem of reconstructing sounds from neural activity, we can finally obtain direct answers to many important questions.
What you will be doing
The work will have three components: building biophysical models of the auditory periphery and using them to simulate neural coding with and without hearing loss; using deep learning to reconstruct sounds from model outputs to determine the effects of hearing loss on peripheral coding; and conducting psychoacoustic experiments to determine how perception of the reconstructed sounds differs with and without hearing loss.
Who you will be working with
The primary supervisor will be Prof Nicholas Lesica. The secondary supervisor will be Prof Stuart Rosen. The project will also involve partners from UCL’s biomedical research centre and hospitals.
Who we are looking for
We are looking for students with a genuine interest in developing a capacity for proper scientific inquiry. While the technical aspects of the project will be challenging, the bigger challenge will be ensuring that the project yields a meaningful conceptual advance. Meeting this challenge will require managing the tension between refining research questions to make them tractable and ensuring that the answers obtained are actually relevant to real-world hearing. Expertise in deep learning is not required, but strong quantitative and programming skills are. Experience with audio, hearing, and/or neuroscience is a plus.