A recent study from MIT suggests that computational models built using machine learning could closely mimic the structure and function of the human auditory system. This discovery could potentially help researchers in designing more effective hearing aids, cochlear implants, and brain-machine interfaces.
In the largest-ever examination of deep learning neural networks trained for auditory tasks, the MIT researchers found that these computational models generate internal representations that share characteristics with representations seen in the human brain when listening to the same sounds. Also, models trained on auditory inputs, including background noise, proved to generate activation patterns closer to those in the human auditory cortex.
These deep neural networks comprise numerous layers of information processing units that can be trained on large data sets to perform specific tasks. The research team, led by Josh McDermott, an associate professor of Brain and Cognitive Sciences at MIT, found similarities between these models’ representation of auditory inputs and activation patterns seen in fMRI scans of human brains listening to the same sounds.
For the purpose of the study, the research team analyzed nine publicly available deep neural network models and also created 14 of their own. Most of these models were designed to perform a single task—such as recognizing words, identifying the speaker, recognizing environmental sounds, and identifying the musical genre.
They found that the models which demonstrated representations closest to those in human brains were those trained on multiple tasks and had received training on auditory input that included background noise. This further supports the hierarchical organization of the human auditory cortex, which involves processing divided into various stages, supporting different computational functions.
This study brings the team closer to their goal of developing a computer model that can predict brain responses and behavior. The enhanced understanding of the human auditory system and the creation of more accurate models bring forth opportunities for advancements in technology to aid auditory impairments and develop better brain-machine interfaces.
