перевод ра. Neural network (перевод ра. Neural network.docx)

Neural network must be trained before it can be used to solve tasks. Training for artificial neural networks is a process of tuning its architecture (structure of links between neurons) and weights of semantics bonds (influencing on coefficients of signals) for effective solution of assigned task. Usually neural network training is based on some selection. During the learning process, which take place according to some algorithm, network must react on input signals better and better.

Turns out, after processing multiple examples, weights of neural network stabilizes, moreover network gives correct answers for all (or so) questions from the database. In this case, neural network is called “trained”. In software realizations could be seen, that during the learning process value of mistakes gradually decreases. After errors number has been reached zero or acceptably low level, the training is stopped and derived neural network is considered to be taught and ready to be used on new data. The degree of teaching sufficiency is defined mainly by requirements of specific task and wished results.