Russian and Chinese physicists have developed a machine learning system that allows for thousands of times faster selection of operating parameters for lasers generating ultrashort radiation pulses. This was reported by the Skoltech press service.
This refers to installations that create attosecond flashes – pulses lasting billionths of a nanosecond. Such lasers are used to study processes inside molecules and materials at a fundamental level.
The main problem is the complexity of calculations. Parameter selection requires modeling the interaction of laser radiation with plasma generated when irradiating metal foil. Each such simulation consumes significant computational resources.
Sergey Rykovanov, Associate Professor at the Skoltech AI Center, noted that the system learns from already obtained data and then quickly evaluates new laser configurations.
The combination of a neural network surrogate model and accurate calculations significantly speeds up the search for promising modes without losing the physical meaningfulness of the result.
As a result, the system sharply reduces the number of complex calculations while maintaining accuracy. This allows for faster identification of optimal operating modes for installations.
The new approach can be applied not only in laser physics but also in other tasks where acceleration of resource-intensive simulations is required.
