Today, academics have shown how by analysing reality and applying data processing techniques we have made progress in precision techniques related to deep learning technologies.
What is the deepest branch of Machine Learning? Researchers of the subject answer: Deep Learning. What is it about? Marvin Minsky, father of Artificial Intelligence and author of the book ‘The Society of Mind’, formulated the theory that a ‘society’ of tiny non-intelligent components, working together, is able to form an intelligent mind. This theory has largely proved true: the techniques we use to connect the incredible number of small parts have been discovered through an infinite succession of trials and errors up to the realisation of neural networks convergence, Machine Learning, Deep Learning, and extended data processing.
According to Osservatori.net, with the right amount of data ‘The system is able to learn the correct representation and solve machine learning problems without the need for data pre-processing, as is the case with traditional Machine Learning techniques.’ In other words, artificial neural networks are exposed to a large amount of data so that they can learn to perform tasks and functions. It is a more sophisticated automatic learning technique that teaches computers to perform a natural function for humans, such as learning.
Among the first examples that we can frame as derivatives of this technology are self-driving cars, which must brake at the stop sign or during their transit must recognise obstacles such as a street lamp or a pedestrian. In addition, as MathWorks explains, ‘It is the key element of voice control in devices such as mobile phones, tablets, TVs, and hands-free speakers.’
In Deep Learning, a computer model learns to perform classification activities directly from images, texts or sounds. Deep Learning models can reach a precision that, at times, could exceed human performance. This is, therefore, one of the main features of Deep Learning which, developments over the years, have made it possible to achieve important results such as the classification of objects in images.
Osservatori.net gives us a detailed photograph of the application areas. In addition to the classification of images it is possible to think of real-time translation, the so-called video captioning, video surveillance, facial recognition, up to medical diagnosis. A practical and recent example of the positive impact of Deep Learning is provided by an article from a learning perspective published on arXiv and taken up by Internazionale. In particular, some computer scientists have used Artificial Intelligence to identify the missing pieces in Mesopotamian tablets. A Deep Learning program trained to read 104 different languages including Semitic ones, including Akkadian from ancient Mesopotamia, which was put to test with ten thousand clay cuneiform tablets engraved 4,500 years ago. The software found missing words or parts of sentences with 89% accuracy and, in some cases, even expanded the possible interpretations of the texts.
Although technologically new, Deep Learning techniques have their roots in the past, precisely as early as the 1980s, but only in the last ten years they have been able to demonstrate their usefulness in a wide range of applications. What made it possible to achieve this result? The available data has increased the development of high-performance parallel computing systems based on GPUs (Graphics Processing Unit) and the optimisation of training methods for neural networks considerably.