Design and Development of a System for Searching of Similar Intellectual Property Objects Using the Neural Network Approach.

The aim of the work is to develop a system for searching for similar objects of intellectual property using a neural network approach. During the development, the methods of building the architecture of neural networks, as well as the methods of their learning, were studied. Learning transfer and fine-tuning methods were chosen as the main ways of building neural networks. Such frameworks for working with neural networks as Keras, Theano and TensorFlow, methods and classes of these libraries were analyzed. In order to train network an adequate amount of time, a platform Colaboratory with a connected grafic processor was used. As a result, a neural network model was obtained with an accuracy of 98% on the test sample. A system that can be used to identify similar images is developed. Its implementation in the processing of incoming applications will significantly reduce the time and resources spent on this work.