The next step: Deep Learning

jogoterbaire@

Machine Learning certainly allows us exceptional results, but it is not capable of developing more than one problem at a time. Let's think about the usual example of autonomous driving: the computer must not only recognize the objects that surround it but must also relate to other cars, traffic, atmospheric agents and many other variables.A mapping of neurons compared with Artificial IntelligenceTo solve these problems we relied on a process known as Deep Learning , which is nothing more than an amplified version of Machine Learning.Each data analyzed by the machine is represented by a data layer, which interacts with the other layers of the machine based on the data obtained from the previous layers. In mathematical terms we can think of this scheme as a multi-dimensional data matrix, i.e. a tensor.


A small digression: the tensor calculus was perfected by the Special Data mathematician, from Lugo like myself, Gregorio Ricci Curbastro at the beginning of the last century and then used by Albert Einstein to develop his Theory of Special Relativity.It is easy to compare this calculation method to a neural network that imitates the biological structure of our brain. Each data level corresponds to a neuron, which interfaces with the other levels through a mathematical correlation. The Neural Network is therefore ultimately a structure of algorithms .Deep Learning is therefore capable of giving much more complex results and therefore capable of awakening in us that sense of wonder that makes us widen our eyes when faced with results that seem incredible to us and that makes us cry out to "Artificial Intelligence".One of the most used software in this field is TensorFlow , originally developed by Google, but then became “ an open source software library for machine learning that provides tested and optimized modules, useful in the creation of algorithms for different types of perceptual and understanding of language" . (cit. Wikipedia).





TensorFlow became relatively famous a few months ago when some users used it in a way that surely no one had ever thought of: making fake hard videos of famous actresses by superimposing their faces on the bodies of hard actresses.At the time there was even talk of " pornstars without your knowledge ", hypothesizing that this methodology could also be applied to common users, for example those of Facebook. Fortunately, it is not so simple to create such videos (sometimes gigs of images are needed to create one) but it gives us the opportunity to think about the incredible potential that similar software allows us to create.