According to the researchers at Los Alamos National Laboratory, neural networks benefit from periods of downtime, just like humans need a good night’s rest. The topic is present at the Conference on Computer Vision and Pattern Recognition.
This is due to the fact that a neural net can become unstable after working for a long period, between powering facial recognition systems, filtering email spam, and other AI works. Thus, the researchers came out with some statement which part of machine learning deserves some time to relax.
At Los Alamos there was a computer scientist named Yijing Watkins, whose team had been studying “spiking neural networks”. Unsupervised learning that proceeds for a long period will lead to the instability of neural set simulations. Then, Watkins and team came out with an assumption which a digital kind of “nap” can help the system return to statis.
“Epochs of sinusoidally- modulated noise that we hypothesize are analogous to slow- wave sleep.” is what Watkins and team describe in their paper in the Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops.
This is to achieve the goal which semblance of artificial sleep, noise had to be injected into the neural network. Team experimented with various types of static, but the final decision is Gaussian noise waves that include an array of various frequencies and amplitudes.
WhatsApp our Team for further assistance
https://go.bigdomain.my/wasally +60 18-767 4068