STOCHASTIC BACK PROPAGATION FOR SCALABLE AND INFERENCE LEARNING

Neural networks and deep learning currently provide the best solutions to many problems in image recognition, speech recognition, and natural language processing. This work presents a novel algorithm based on stochastic back propagation rules through stochastic variables. The proposed algorithm is scalable for inference and learning. In order to demonstrate the approach 2 networks were created with 100 hidden layers and trained using training set of different sizes. The proposed approach has been validated using MSIT data set and also though comparison of results available in the literature. The comparison is made in terms of average RMSE value. The result demonstrates the improved performance of the proposed approach.