Scientists of the University of Wyoming (USA) found that the decision-making, expressive speech and arbitrary movements correspond to the recurrent neural networks in the brain. Artificial analogues are used to solve problems such as language translation and recognition of objects. This is reported in an article published in the Cell Reports magazine.
Recurrente neural networks (RNN) are an oriented graph where the ribs between nodes have a direction. In the mammalian brain, RNN is the center that receives input signals from the areas of the brain responsible for emotions, and the output is sent to the motorcore that is responsible for movement. Thus, the neural network is responsible for behavioral strategies, and artificial intelligence on it can be used to solve real problems, such as language translation, processing of the natural language, speech recognition and subtitles.
Researchers analyzed the brain of various lines of genetically modified mice, marking special types of neurons with fluorescent proteins, as well as adjusting the activity of individual nerve cells. It turned out that structures similar to artificial recurrent neural networks are located in the frontal lobe of the brain are less complex than the researchers assumed, and are mainly unidirectional.
Results will help researchers to apply computer simulation to predict how the brain encodes short-term memory, and understand how to use this mechanism.