“The DiCarlo lab finds that a recurrent architecture helps both artificial intelligence and our brains to better identify objects ” MIT news
From an engineering perspective, we should understand why the brain needs recurrent architectures, when we need them, and how we can operationalize this procedure into our deep neural networks.
This work definitely has started the first fundamental step to reach our goals. However, as I mentioned, we still need to know more profound about this research, such as the precise procedures and how many neurons involved.
Object recognition in our brains is not working alone instead of links with high-level cognition, such as emotion and memory. They are much likely cooperating with the visual cortex in object recognition. Thus, to overcome the challenges of object recognition in artificial intelligence, we have quite a lot of work to do indeed.
Paper: Evidence that recurrent circuits are critical to the ventral stream’s execution of core object recognition behavior -Authors: Kar, K, Kubilius, J, Schmidt, K, Issa, EB, DiCarlo, JJ Nature Neuroscience