Cross-fertilization of neural network/machine learning theory and R&D in physiology Clearly, the development of neural networks has been based, at least initially, on those occurring in actual animals, human or otherwise. Conversely, how have neural-network or machine-learning models/theories informed/directed/aided R&D in physiology (that of humans or other animals) or biomedical engineering, e.g., in development of prosthetics or exoskeletons for humans or control circuitry for other organisms or the mapping of actual neural networks in the human brain?
References to survey articles would be especially useful.
 A: A recent example is: The Code for Facial Identity in the Primate Brain - Chang, Tsao, where the authors provide a very convincing explanation of how the primate brain encodes human faces using only 200 neurons. Effectively, the brain has it's own version of of principle component analysis along with a relatively simple neural network.
Overall, the analogies between neural networks and the cortical brain are largely specious. While initial research was perhaps inspired by the way the brain works, the actual structure of cortical neurons is completely different from neural networks, one distinction being that neurons operate using spike trains (there are spike-train neural networks in machine learning as well). There's a nice talk by Hinton here, where he discusses a few hypothesis for why the human brain uses such spike trains (spoiler: it's to prevent overfitting).
Finally there's a nice overview of brain function with analogies to machine learning here: Dendritic Learning - London, Hausser
