Artificial intelligence techniques have experienced rapid development in recent years and have led to extremely significant gains both in terms of speed and quality in many applications. They are thus widely deployed when large amounts of data are available: they make it possible to accurately recognize different characteristics and offer relevant predictions, whether on social networks, in video games or as assistance in piloting an autonomous car.
These artificial intelligence techniques are based on virtual neurons networked and which are able to learn from a few rules or indeed from data collections. In a recent contribution , researchers from the Solitons Lasers and Optical Communications team of the ICB laboratory, in collaboration with Aston University, implemented these methods for solving problems in applied physics such as the prediction of changes affecting light traveling through an optical fiber. If we know how to solve this question numerically effectively, the necessary computing times become considerable when several tens of thousands of different combinations are to be tested. Once a learning phase is completed, a neural network is able to deal with complex problems that would have taken more than a day with traditional computer programs in just a few seconds. It therefore becomes possible to identify almost instantly the parameters to be implemented to obtain a specific light pulse. Neural networks therefore constitute a precious help in imagining new technical solutions in order, for example, to manipulate signals of a few thousandths of a billionths of a second at very high rates and in an entirely optical manner.
Potential applications are not limited to the field of fiber optic telecommunications. They also prove to be relevant in the design of fiber laser sources . Colleagues from the University of Franche-Comté are successfully exploring the use of neural networks in the context of extreme prediction in optical systems.