The Russian company Cognitive Pilot has devised a method to speed up the process of training neural networks by up to 20 times, the Sputnik news agency reported here today.

According to the report, the specialists of the company, which is a joint subsidiary of Sber Bank and the software corporation Cognitive Technologies, created «an automatic mechanism for analysing and correcting the mistakes that a neural network makes during training».

The new technology, Cognitive Neural Network Hospital, «makes it possible to reduce the training time by 15-20 times» and increase its performance by 40 per cent.

It first detects the weakest points in a neural network and then new data that is not identical but similar to those errors, which it analyses and classifies into groups.

Cognitive Pilot CEO Olga Uskova noted that this technology provides the opportunity to radically accelerate the development of artificial intelligence systems for autonomous driving, which «ultimately provides users with the continuous use of the most advanced and perfect solutions».

Such technologies, she added, pave the way for «a promising new market that could be worth several billion dollars in the next five years».