Self-Organized Auditory Balance

People can sense the location of a sound source by the slight difference of the arrival time. Meanwhile, due to the acoustic shadowing effect of the head, the ears can pick up the difference in amplitude. Thus, it is expected that the auditory cortex must process the signals from hair cells with equal weighting so that the amplitude difference can be detected. As I was reading the book Auditory Neuroscience, a question popped into my brain – how do people grow balanced hearing between both ears?

I decided to explore the answer by numerical simulations. Treating the connections between ears to the auditory cortex as the simplest neural network, this project utilizes generalized Hebbian learning algorithm to simulate how the connection weights change with the external stimuli. If the ears receive positively correlated white noise as input, the numerical simulations show that the auditory balance is self-organized with equal connection weights. On the other hand, the weights trained by non-correlated white noise will fluctuate and the auditory balance is not guaranteed. The numerical simulations show that the auditory balance is a robust consequence from unsupervised learning due to correlated auditory inputs – very amazing!