DYNAMIC STRUCTURE NEURAL MACHINE FOR SOLVING PREDICTION PROBLEMS WITH USES IN MACHINE LEARNING

摘要

This invention discloses a new and novel methodology which can be used to solve multiclass classification problems in an automated way. It describes a novel neural network architecture "Dynamic Structure Neural Network (DSNN)", a novel automated learning method "Dynamic Structure Neural Learning (DSNL)" for training DSNN models and a product "Dynamic Structure Neural Machine (DSNM)" which is a computer-implementation of DSNN and DSNL for solving multiclass classification problems, such as, Medical Diagnosis, Face Recognition, Sentiment Analysis, Speech Recognition e.t.c. The system and method given in this invention analyzes any (structured, semi-structured or unstructured) type and form of data that can be vectorized. The novelty of this method is the architecture of the DSNN model and automated learning method DSNL that simultaneously determines the number of hidden layers, number of processing units (or neurons) in each hidden layer hidden layer and their parameters (weight and biases).