Towards Machine Learning for Enhanced Maths Teaching to the Blind

This summary refers to a complete research paper. In 2010, the World Health Organization (WHO) estimated that 19 million children under the age of 15 were visually impaired, with about 39 million blind and 246 million people with severe or moderate vision loss. These numbers suggest the potential size and potential impact of the visually impaired on a day-today basis. The situation has seen some improvement in recent years with increasing access to formal education, from primary school to higher education. As a result, the demand for innovative, technology-based assistance tools that enhance the user experience and the quality of education has increased. This work takes a step towards bridging a technological gap with the design and validation of a system of artificial neural networks, called as Deep Neural Networks (DNNs), capable of identifying the main Cartesian curves of the mathematics curriculum. The Cartesian set comprises 6 degrees of rational algebraic curves, in addition to a total of 42 conic curves, for this paper will be presented two sets of curves, parabolas and ellipses. The development of this gave some methodological steps. First, relevant artificial neural networks were investigated considering the needs of the user and the space of the problem - computer efficiency, loss rate and precision of the generated models used as selection metrics for each neural network tested. The selected network models were InceptionV3, MobileNETV2, VGG16, and VGG19. With average accuracy of 94% and 20% of mean loss, the selected network for the application was VGG16.