Applied deep training. Approach to understanding deep neural networks based on the case of cases

Extended topics of deep learning are affected: optimization algorithms, tuning hyperparameters, screening and analysis of errors, strategy for solving typical problems during training of deep neural networks. Simple activation functions with a single neuron (Relu, Sigmoid and Swish), linear and logistics regression, Tensorflow library, the choice of cost function, as well as more complex neural network architectures with numerous layers and neurons are described. Debugging and optimization of expanded methods of weming and regulatory, setting up machine learning projects focused on deep training using complex data sets is shown. The results of the analysis of errors of the neural network with examples of solving problems arising due to dispersion, displacement, retraining or disparate data sets are given. For each technical solution, examples of solving practical problems are given