Science for decades has been trying to discover the secrets of the work of the human brain, so often there are cases where scientists decide to create a clone or their version of the brain. The Russian Science Foundation has just announced that one of these experiments ended with great success.
A group of scientists from the Petrozavodsk State University created a neural network of the simplest artificial neurons and scientists have learned to recognize geometric figures and images, the press service of the Russian Scientific Fund reported.
Modern computers work on the basis of the simplest discrete logic – their elementary memory cells and computer modules can only perceive and process only zeros and units. It provides them with unlimited possibilities for mathematical calculations with a sufficient number of elements and elements.
Scientists have long known that the brain functions in a different way. Unlike semiconductor transistors, our neurons can simultaneously perceive many different signals, simulate them in a complex way, and alter their sensitivity to the individual assemblies of such impulses.
The neural network is actually an artificial chain of neurons and is therefore very difficult to create. The chain of neurons is too complex to display in full form, so scientists often decide to simplify their construction, which makes insufficiently good pendants of the human or the brain of other living things. In addition to all other challenges, such feats can be made only with the help of the most powerful supercomputers in the world.
Because of that Andrei Velichko, as well as other physicists and scientists from around the world, have long been trying to create artificial neurons. Recently, Russian scientists created artificial neurons made up of thin films of vanadium-dioxide, whose resistance depends on their temperature. For example, if a sufficient amount of electricity is missed through such a neuron, it will warm up and its resistance will shrink about 10,000 times, causing it to start transmitting more current. This will lead to a cooling of the artificial neuron, its return to its initial state and almost complete interruption in the movement of electricity. As a result, this film will oscillate, emitting energy through neurons in a similar way to humans and animals.
As Velichko and his colleagues discovered, a total of 11 similar structures combined in a network consisting of two input and output nodes and nine processing neurons can learn to recognize simple geometric shapes that can be placed in a square size of 3×3 pixels.
These experiments have shown that artificial vanadium-based neurons can record and process multilayer signals, just like natural neurons, which allows them to be used as their full-bodied pendants.