Thoughts out loud
Recognition of mental commands with an accuracy of up to 85%
The prostheses controlled with the help of neurointerfaces are an advanced technology that can improve the lives of people with damaged spinal cord and other mobility problems. Such devices, although experimental series, are manufactured. But now they cannot be used "out of the box", without pre-configuring for a specific person. During the training, the computer must learn to understand silent orders, and the person must give them. This is a rather long and time-consuming process. The question arises — how similar are the signals transmitted by the brain when the same command is given by different people? Is it possible to teach a computer to "understand" the orders of any person?
The study, the materials of which are published in the journal Sensors, answers these questions.
"The main result was the transfer of learning from one person to another in the task of recognizing mental commands using the neurointerface. That is, a model trained on some people is able to recognize the mental commands of others with some accuracy," he said Sergey Markov, one of the authors of the work.
During a series of experiments, patterns corresponding to nine words of the Russian language, including directions (forward, backward, up, down), commands (take, stop and release) and a request for help, were distinguished from electroencephalography data (40-channel EEG, registration of a signal with a frequency of 500 Hz) of healthy people. Neural networks of various configurations were used for this, ResNet18 + 2GRU gave the best results. It was possible to achieve recognition with an accuracy of 84.51%. Despite a number of problems, it shows the presence of common patterns in the brain activity of different people and the ability to recognize such patterns using a non-invasive method and use data to control computerized devices.
Portal "Eternal youth" http://vechnayamolodost.ru