A team of specialists proposed a fundamentally new training algorithm
What good does it do?
Every week we learn about all the new feats of computer programs that already know how to put medical diagnoses, draw no worse than Rembrandt, sing or generate text.
But no matter what anyone says, people are good in many ways: many of us are infected with the thirst for discovery, we know how to play chess and solve complex political issues. Therefore, scientists do not stop there and try to make artificial intelligence even smarter and more advanced.
So scientists from Purdue University have developed a technology for the use of magnetic radiation for programming and training robots, unmanned vehicles and aircraft.
It will allow artificial intelligence to more effectively receive and process information, summarize the data and independently draw conclusions about the surrounding objects, identify their properties and differences.
“Our neural networks mimic certain activities of the human brain and perform calculations by connecting synapses of neurons like the human brain,” said Kaushik Roy, Professor of computer and electrical engineering at Purdue University.
The team of specialists proposed a fundamentally new training algorithm based on the biological processes of regulation of the strength of connections between neurons, Spike-timing-dependent plasticity, which were found experimentally in the hipocampus of a laboratory rat.
Magnetic radiation in the technology is used to control the state of magnetization of neurons, which allows you to influence the energy efficiency II, processing speed and compression of data.
“We have created a simpler neural network that can be used to solve complex problems and optimize processes such as charting and routing,” Roy said.