Hopfield's Neural Network is More Powerful When it Sleeps and Dreams!

The article discusses how allowing Hopfield's neural network to have a proper sleep and dream enhances its performance. Scientists drew inspiration from sleeping mammals during the development of this neural network.

Essentially, it revolves around a biological process known as synaptic homeostasis, which involves the "pruning" of excess synaptic connections that the brain has built up during the day. The aim is to prevent the brain from becoming overloaded with unnecessary information, thereby enabling it to function more efficiently.…

In the article, you will read about Adriano Barra from the University of Salento in Italy, who, along with his colleagues, programmed the Hopfield neural network to be capable of sleeping, inspired by mammalian brains.

The results of experiments with the sleeping network appear to be fascinating. A neural network that does not sleep has a maximum capacity corresponding to the number of bits stored per synapse, alpha = 0.014. However, when the neural network was allowed to sleep, its capacity reached the theoretical limit, at alpha = 1. If I am calculating correctly, this means that with the same number of neurons, sleep can increase the amount of memorised information by up to 70 times!

Sources: http://www.osel.cz/10371-nova-umela-inteligence-je-vykonnej…

Neural Networks online 29. 1. 2019, Science Trends 14. 2. 2019.