Neural network learning algorithms with smoothing

The aim of the work is to investigate the impact of the use of weight smoothing in the neural network model during training on the quality of prediction and the time needed to train it. Certain types of calculating moving averages were used for smoothing. The proposed smoothing algorithms were tested on the VGG19, Wide Residual Networks 28x10 and Densenet 100x12 models. The aim of the work is to examine the prediction results of models with smoothed weights in comparison with models trained without the use of weight averaging, also taking into account the time needed to train the model.