Novosibirsk State University, Russia
Alexey Kokhanovskiy is a researcher of department of Laser Physics and innovative technologies at Novosibirsk State University. His specialization is developing new schemes of mode-locked fiber lasers and investigating nonlinear dynamics of pulse formation inside the laser cavity.
Currently he is involved in a project dedicated to application of machine learning algorithms in fiber lasers. In his talk, he will be describing how commonly encountered challenges involving design and engineering of mode-locked fiber lasers can be solved by artificial neural networks, random forests and other machine learning approaches.
Mode-locked fiber laser controlled by machine learning algorithms
One of the modern trends in development of mode-locked fiber lasers is a focus on precise adjustment of temporal and spectral properties of optical pulses at the expense of increasing complexity of system design. Complexity imposes extra requirements on control and feed-back systems design in terms of maintaining accuracy. s. In this context, machine learning (ML)-based approaches offer a nonlinearity-friendly, efficient and flexible alternative to the classical control techniques. First part of this talk describes a possibility to manage temporal and spectral parameters of pulses generated from fiber 8-figure laser. It will be demonstrated that self-tuning adjustment of the two independent gain levels in the laser cavity and application of various objective functions enabled pulse generation with parameters on-demand. Second part is devoted to a problem of implementation a feed-back system linking the laser performance and variable laser cavity parameters. Application of machine learning algorithms allows cutting down a number of necessary measurement equipment to get basic parameters of ultrashort pulses. Processing dispersive Fourier transformed pulse train is enough to determine spectral, temporal and coherence properties of the pulses by solving regression problem via artificial neural network or extremely randomized trees.