The use of neural networks for solving problems of various orientations has become quite popular recently. Including neural networks are used in learning tasks with reinforcement as a management system, in which learning occurs through interaction with the environment. The article is devoted to the analysis of the control problem of an inverted pendulum, for which such aspects as redundancy of the used neural network are investigated, i.e. one of the tasks is to search for a more optimal form of a neural network; uniqueness of the solution. In addition, the article explains and justifies the choice of the number of layers in the neural network used. In the work it is established that the neural network used can be brought to a third of the least so that the network will continue to hold the reverse pendulum, which indicates its redundancy. There was also the interpretation of synthesizing neural network, which the authors have not previously encountered, namely, the neural network is a classifier, which identifies the hidden layer characteristics for the peremescheniya trolley, and the output layer is an aggregator, which is featured on the received outputs the control signal. This interpretation allowed to reasonably explain the size of the hidden layer of the neural network and, as a result, to reduce the number of neurons from 128 to 16, which can be critical for embedded systems, and also reduce the learning time from 2600 epochs (on average) to 1300.

1. PyTorch examples. Available at: https://github.com/pytorch/examples (accessed 25.04.2018).

2. Bishop C. Pattern recognition and machine learning. Information science and statistics. New York, Springer, 2007. 738 p.

3. Richert W., Coelho L. Building machine learning systems with Python. Birmingham, Packt Publ., 2013. 290 p.

4. Hastie T., Tibshirani R., Friedman J. The elements of statistical learning: data mining, inference, and prediction. 2^nd ed. Springer series in statistics. New York, Springer, 2013. 745 p.

5. Lantz B.Machine learning with R. Birmingham, Packt Publ., 2013. 396 p.

6. Mohri M., Rostamizadeh A., Talwalkar A. Foundations of machine learning. Adaptive computation and machine learning series.Cambridge, MA, MIT Press, 2012.432 p.

7. Conway D., White J.M. Machine learning for hackers. Sebastopol, CA, O'Reilly, 2012.324 p.

8. Welcome to the Deep Learning tutorial.Available at: http://deeplearning.stanford.edu/tutorial/(accessed 25.04.2018).

9. Haykin S. Neural networks: a comprehensive foundation. New York, MacMillan Publ., 1994. 1104 p.

10.Romannikov D.O. O preobrazovaniiseti Petri v neironnuyu set' [On the transformation of Petri nets in neural network]. SborniknauchnykhtrudovNovosibirskogogosudarstvennogotekhnicheskogouniversiteta – Transaction of scientific papers of the Novosibirsk state technical university, 2016, no. 4 (86), pp. 98–103.

11. Goodfellow I., Bengio Y., Courville A. Deep learning.Cambridge, MA, MIT Press, 2016.

12. Krizhevsky A., Sutskever I., Hinton G.E.ImageNet classification with deep convolutional neural networks. Proceedings of the 25th International Conference on Neural Information Processing Systems: NIPS 12, Lake Tahoe, Nevada, 2012, pp. 1097–1105.

13. Graves A., Mohamed A., Hinton G. Speech recognition with deep recurrent neural networks. Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2013, Vancouver, Canada, 2013.

14. Deng L., Hinton G.E., Kingsbury B. New types of deep neural network learning for speech recognition and related applications: an overview. Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2013, Vancouver, Canada, 2013.