The story of the first neural networks creation takes its origin in the 40th of the last century. In 1949, Hebb spoke about the organizational principles of connecting brain neurons and their interaction. Later, work ideas and terminology for artificial neural networks were taken from biology. Coincidence end there.
Until recently, the construction of an artificial intelligence was impossible due to personal computers insufficient capacity and the lack of super-large databases necessary for training and optimizing the neural network progress. Vinberg began as an enthusiast back in 2003, as a research theorist, but today he is a recognized expert with practical experience. Andrews specializes in multi-layered neural networks. His participation in the advisor role brings the success of our company.
There are about 100 billion neurons in the human body – nerve cells responsible for the transmission, storage and processing of electrical and chemical signals. Impulses come to them from external and internal receptors, as well as other neurons. For example, the object we see is read by the retina. Then the image in the electrical impulses form enters the cerebral cortex, where it is “intercepted” and analyzed by neurons.
How does it work with artificial intelligence? In artificial neural networks, this node is called an artificial neuron or processor. For example, a simple, single-layer network can distinguish a circle from other shapes. The system receives a large amount of incoming data in the form of various shapes images, as well as the circle we need. The incoming layer of neurons analyzes and classifies the parameters of the incoming data and returns 0 if the value is less than a certain threshold of parameters, and 1 if the value corresponds to certain parameters. In plain language, a simple single-layer neural network tells us “no” or “yes” when parallel data processing is in progress.
The neural network is able to link input data and data received at the output. For example, a square may be black, large, oblique, but it must have four corners. Thus, by finding connections and comparing data with threshold values, artificial intelligence is able to learn. This reduces the errors number that system can make when starting work with new information. A simple network has input and output artificial neurons layers.
Multilayer neural networks also have several layers, called hidden ones, and allow us to conduct data analysis sequentially, from level to level. For example, we can give our shape additional features: color, size, location relative to the axes and other features, and teach the neural network to determine it with a minimum errors degree. It was not by chance that Vinberg proposed a scheme for using multi-layer neural networks to solve the problems that Intelligent
Medical Developments company facing
The ability to process a large number of input parameters and data allows multilayer neural networks to become an indispensable assistant in solving complex problems. For example, diagnosing the skin lesions depth in case of burns during mass hospitalization. Diagnostics of somatic diseases is made on the basis of the clinical picture, examination data, analyzes and the collected patient’s life history. It is necessary to identify the main symptoms, classify the information received and make a conclusion. The main task of Andrews today, together with doctors and other project participants, is to transfer a similar algorithm to the developed neural network model.
Medical neural networks today determine pathology based on the input of visual information, therefore, are not applicable in all cases. However, skin diseases and injuries, oncological pathologies, eye diseases and heart attack risk determination by retinal images are modern artificial intelligence possibilities in medicine.
What happens tomorrow? Tomorrow will come only with such enthusiasts as Andrews Vinberg, who are not afraid to discover new technologies and put them to the betterment of mankind.
Editor of IMD News