SUMMARY
Neural activity from animals is often used as a proxy for the human brain. However, due to distinct environmental pressures, the relevance of perceptual systems described in animal models can be unclear. This problem is accentuated when animal physiology and human behaviour are not in complete agreement, as is the case for binaural hearing-in-noise. As a means to bridge this gap we reverse-engineered artificial neural networks from binaural psychophysics. By comparing in silico “physiology” in neural networks with in vivo animal data, we were able to make inferences as to the basis of binaural perception in humans. We observed the emergence of highly specialized solutions to account for low frequency sound detection. Artificial neurons developed a sensitivity to temporal delays that increased hierarchically and were widely distributed in preference. Network dynamics were consistent with a cross-correlator, comparable to the type reported in animal physiology. Our results attest to the likely prominence of this neural mechanism in human biology. Moreover, this is a primary demonstration that deep learning can infer tangible neural mechanisms underlying auditory perception.
Competing Interest Statement
The authors have declared no competing interest.