Despite the problems we have illustrated, Deep Learning systems have made enormous evolutionary steps. They have improved a lot in the last five years, above all of the vast amount of data available but all for the availability of ultra-performing infrastructures (CPU and GPU in particular ). As part of Artificial Intelligence research, machine learning has enjoyed considerable success recently, enabling computers to outperform or approach corresponding human performance in areas ranging from facial recognition to speech and language recognition.
On the other hand, deep knowledge allows computers to take a step forward, in particular, to solve a series of complex problems. Today there are use cases and application areas that we can also see as “ordinary citizens” who are not tech-savvy. From computer vision for driverless cars to robotic drones used for the delivery of packages or even for assistance in emergency cases (for example, for the delivery of food or blood for transfusions in an earthquake or flooded areas or in areas that must deal with epidemiological crises, etc.). Speech and language recognition and synthesis for chatbots and service robots.
Facial recognition for surveillance in countries like China; image recognition to help radiologists find tumors on X-rays, or to help researchers identify disease-related genetic sequences and identify molecules that could lead to more effective or even personalized drugs; analysis systems for predictive maintenance on infrastructure or plant by analyzing the data of the IoT sensors; and again, the vision of the computer that makes the Amazon Go supermarket without a cashier possible. Looking instead at the types of applications (intended as tasks that a machine can perform thanks to Deep Learning), the following are the most mature ones to date:
- automatic coloring of black and white images (for the neural network, it means recognizing edges, backgrounds, details, and knowing the typical colors of a butterfly, for example, knowing exactly where to place the correct color);
- automatic addition of sounds to silent films (for the Deep Learning system, this means synthesizing sounds and placing them correctly within a particular situation by recognizing images and actions, for example, by inserting the sound of the pneumatic hammer, the breaking of the asphalt and the substrates of a busy city street in a video of workers breaking the asphalt with a jackhammer);
- simultaneous translation (for the Deep Learning system, it means listening and recognizing natural language, recognizing the spoken language, and translating the meaning into another language);
- classification of objects within a photograph (in this case, the system can recognize and classify everything it sees in an image, even a very complex one where, for example, there is a background landscape, for example, mountains, people walking along paths, grazing animals, etc.);
- automatic generation of handwriting (there are already Deep Learning systems capable of using human handwriting to write, even learning the estimates of human handwriting and imitating it);
- automatic generation of text (these are systems that have learned to write correctly in a specific language respecting spelling, punctuation, grammar, and even learning to use different writing styles depending on the output to be produced, for example, a journalistic article or a short story );
- automatic generation of captions (in this case, the recognition of the images, the analysis of the context, and the ability to write allow a system to automatically write the captions of an image, perfectly describing the scene);
- automatic game (we have learned to understand the potential of a system capable of independently learning how to play a specific match thanks to DeepMind – now part of Google – which has developed a Deep Learning system – AlphaGo – which has not only learned playing the very complex Go game but also managed to beat the world champion, human).
Traffic Sign Detection (TDS) is a feature found on many new cars that allows you to recognize road signs. It is a machine learning application that uses convolutional neural networks and frameworks such as Tensorflow.
In early 2021, a new AI methodology will be applicable in film production. The new approach uses different parallel and series deep learning tools (VGG16, MLP, transfer learning). It uses other datasets (images with various features highlighted) to classify film shots accurately. And thus allow a professional and operational use in the filmmaking process or the indexing of streaming contents. Let’s talk about AI applied to image processing.
Great expectations are focused on quantum computing . A computing paradigm that requires new hardware, new algorithms, and new solutions. The main feature is the ability to greatly simplify the solution of some problems, reducing their exponential complexity. An example: is the determination of the factors of a number, at the base of many cryptographic issues and on which many IT security applications are based.
A study conducted on holograms in 2021 by researchers from the Massachusetts Institute of Technology (MIT) proved how, thanks to a new deep learning technique called “tensor holography,” it is possible to generate a holographic video instantly, exploiting the present computational capacity on a simple computer. The peculiarity is that of being composed of trainable tensors, which can learn how to process visual and depth information similarly to what the human brain does.
Also Read: IOS15: HOW TO USE THE LIVE TEXT FUNCTION