Humans and their cousins, the apes, have been exceptional in
facial recognition. Their life depended on it for the ability to recognize a friend
from a foe was crucial for survival. Recent research from Caltech (1) appears
to get closer to the neural code for face recognition. The trick appears to be
specialization with singular neurons, focused on specific features. The idea
appears to validate recent advances in deep learning for image recognition and it
could provide further impetus to the acceleration of artificial intelligence.
The idea that single neurons encode specific features has
been tantalizing for deep learning enthusiasts. It allows scalability in deep
neural networks with increasing specialization in layers. The single feature
specificity at the neuron level and its ability to make binary decisions
provide further evidence of micro compartmentalization and voting based
decision-making. Complementarity, communication and cooperation appear to
dominate in goal seeking and that has implications for future research in
computerized image recognition and more broadly, artificial intelligence.
The neural code developed by primates, an efficient and
massively parallel processed algorithm that helped them survive and evolve,
could be replicatable. Hopefully, those who get hold of this will use it
wisely.
(1) https://www.scientificamerican.com/article/how-we-save-face-researchers-crack-the-brains-facial-recognition-code/
No comments:
Post a Comment