Mathias Dietz: Investigating Binaural Hearing with Deep Neural Network Models trained for Sound Localization
| When |
Feb 19, 2025
from 12:15 PM to 01:00 PM |
|---|---|
| Where | Bernstein Center, Hansastr. 9a, Lecture Hall. |
| Contact Name | Gundel Jaeger |
| Add event to calendar |
 vCal iCal
|
Abstract
Deep neural networks trained to localize sounds using simulated binaural auditory nerve input have emerged as new models of human sound localization. Despite never being fit to human data, these networks reproduce patterns of human behavior in several classic psychoacoustic experiments, suggesting they learn to use similar cues as humans do. Here, we critically examine similarities and discrepancies between deep neural networks on the one side and human behavior and mammalian physiology on the other.
Most discrepancies appear to be due to different excitatory-inhibitory interactions in real vs. artificial systems. The ability to generalize from trained sound localization to untrained headphone psychophysics supports other recent reports that deep neural networks are a promising tool for studying sensorineural processing.