A study from MIT has suggested that machine-learning computational models can help design more effective hearing aids, cochlear implants, and brain-machine interfaces by mimicking the human auditory system. The study was based on deep neural networks which, when trained on auditory tasks, create internal representations similar to those generated in the human brain when processing sounds. The models most closely resembled the human auditory system when they were trained on more than one task and on auditory inputs that included background noise.
“This study represents the most comprehensive comparison of machine-learning based models to the auditory system thus far. The study suggests these models could be steps in the right direction, providing clues on what makes them more accurate models of the brain”, says Josh McDermott, an associate professor of brain and cognitive sciences at MIT, a member of MIT’s McGovern Institute for Brain Research and Center for Brains, Minds, and Machines, and the senior author of the study.
The study supports the idea that the human auditory cortex has a hierarchical organization, in which processing is divided into stages that support different computational functions. The researchers found that representations generated at the earlier stages of the model more closely resemble the primary auditory cortex, while later stages more closely resemble the regions beyond the primary cortex. Hence, models trained on different tasks better replicated different aspects of audition.
This research opens possibilities for developing models that can reproduce human brain responses. Beyond contributing to scientists’ understanding of brain organization, such models could also be used to improve hearing aids, cochlear implants, and brain-machine interfaces. “We believe reaching our goal of having a computer model that can predict brain responses and behavior will unlock many possibilities,” McDermott says. The research was funded by various organizations including the National Institutes of Health, an Amazon Fellowship from the Science Hub, and a Department of Energy Computational Science Graduate Fellowship.
