A new type of neural network developed at the University of Michigan, was able to guess the words before they were uttered. Let these “psychic forces” there is nothing truly supernatural, the development can significantly improve the efficiency of artificial intelligence systems.
Artificial neural network is a mechanism working on the same principle that networks of nerve cells of a living organism, that is, by and large, in many ways resembling the human brain. Such networks are not programmed, but learn, that is, “independently” identify complex relationships between input and output data and carry out on this basis generalization. This approach is considered the most promising for the creation of artificial intelligence.
Scientists have created an algorithm to reduce the time required for neural network training, and to increase the amount of information that it is able to master. To do this, the researchers used memristors are passive elements in microelectronics, capable of changing its resistance depending on flowing an electric current through it. Reportedly, the new approach is also much more economical than used today. During the test, the neural network consisting of only 88 memristor, was able to decipher handwritten numbers with accuracy, reaching 91 percent — a more “classical” neural network would require thousands of nodes simulating the neurons.
Scientists believe that in the future, their development could be used for speech recognition and predictive analysis. As noted, the neural network could guess what a person is going to say, “in the middle”, just as often occurs in ordinary conversation.
The study was published in the scientific journal Nature Communications.
Recently, researchers from Russia and Spain have taught a neural network to predict corruption.
The most interesting day in the “MK” in one evening newsletter: subscribe to our channel in the Telegram.