MIT researchers have found that computational models derived from machine learning, designed to mimic the human auditory system, have the potential to improve hearing aids, cochlear implants, and brain-machine interfaces. They are moving closer to this goal by using these models in the largest study yet of deep neural networks trained to perform auditory tasks. The study showed that these models generate internal representations sharing properties seen in the human brain when listening to the same sounds. It revealed that models trained on auditory input with background noise closely mimic the activation patterns of the human auditory cortex.
Deep neural networks are capable of mediating behaviors on a scale that previous models couldn’t. The researchers analyzed nine available deep neural networks and created 14 of their own. They were trained to perform tasks such as identifying speakers, recognizing words and environmental sounds, and identifying musical genre. The model representations exhibited similarities with the internal processes of the human brain. The closest matches were models trained on more than one task and those trained with background noise.
The study also suggested that the human auditory cortex has some degree of hierarchical organization, where processing is divided into stages that support distinct computational functions. Models trained on different tasks performed better at reproducing different aspects of audition, supporting the claim that task-specific optimization selectively explains specific tuning properties in the brain.
The research team plan to use these findings to develop more accurate models for replicating human brain responses. These could potentially enhance design processes for improved hearing aids, cochlear implants, and brain-machine interfaces. The aim is to create a computer model that can predict brain responses and behavior, unlocking new possibilities in the field.
This research was funded by several institutions, including the National Institutes of Health, the Amazon Fellowship from the Science Hub, the American Association of University Women, and the Department of Energy Computational Science Graduate Fellowship.
