Advances in recognition.
An Optimized Solution for Face Recognition
MIT News.Jennifer Michalowski, April 6, 2022
Neuroscientists at the Massachusetts Institute of Technology (MIT) found that a deep neural network taught to identify faces and other objects assigns specific components to perform these tasks. Former MIT researcher Katharina Dobs said any system trained to recognize faces and sort objects would hypothetically mimic the brain's approach of segregating facial and object processing. Dobs compiled scores of images of faces and objects, without telling the network which was which. In learning to recognize them, the network structured itself into an information-processing network that committed specialized units to face recognition, relying on general vision processing machinery in the early stages, and on face-dedicated components in the final steps. MIT's Nancy Kanwisher said the research addresses why the brain has a separate face recognition system, "because that is what an optimized solution looks like." .... '
No comments:
Post a Comment