Alexnet has arisen from teamwork carried out in a small Canadian university. Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton have developed a neural network capable of recognizing images with unprecedented precision. Their system used three technologies: deep neural networks, the imagenet database and graphic processors (GPU).
Their performance was remarkable. The network could properly classify images in 1,000 different categories, from strawberries to school buses, including dog breeds. Yann Lecun, expert in computer vision, described this progress as ” unequivocal turning ».
The system was radically different from previous approaches. Where the old methods required a manual programming of the characteristics, Alexnet automatically learned to recognize models. From simple outline to parts of complex objects, the network developed its own understanding.
Clearly, a neural code is a computer program designed to learn and adapt as our brain does. As software that does not follow rigid instructions, but which can understand models, recognize forms and improve its performance over time.
Unlike traditional programs that exactly do what has been programmed to them, a neural code works as a network of interconnected neurons. These artificial “neurons” receive information, analyze it, and transmit signals, a bit like neurons in our brain.
If we show this type of code thousands of cat photos, he will gradually learn to recognize what a cat is, without being explained everything in advance. It will identify the common characteristics: ears, a tail, a certain shape of muzzle.
These neural codes are used in many areas: voice recognition, automatic translation, detection of bank fraud, or even to help with medical diagnosis. It is a technology that allows machines to “think” more flexible and intelligently. We understand better, therefore, why it is this system that gave birth to all the AI applications that we know and use today.