The rise of machine learning, or artificial intelligence has revolutionized many fields of technology. The rapid progress in the development of both new algorithms and increased computational capabilities have allowed for the training of large neural networks. Not only in the more notable applications such as large language models and diffusion models, but the use of machine learning techniques in science is becoming more and more common to constructing models for complex dynamics.
Ultrafast optics is a particular field of science that could benefit significantly from the use of machine learning techniques.
“Light-matter interactions in optical fibres are highly sensitive to small variations in the parameters of the laser source and optical fibres. This makes machine learning -based approaches very attractive for the development of broadband fibre-based laser sources”, says Lauri Salmela.
In his dissertation, Salmela developed new computational schemes to model ultrafast pulse propagation. One of the key innovations in the dissertation is the use of neural networks for simulating the pulse propagation dynamics. When a high-power laser pulse propagates in an optical fibre, both the temporal and spectral (color) properties of the pulse are changed. A spectacular demonstration of this process is the supercontinuum generation where an initially single-color laser spectrum expands to cover a large span of colours, up to infrared radiation.
“Neural networks can very efficiently learn to model this process. As the predictions of the neural network are essentially instantaneous, the machine learning approach offers a significant advantage in terms of simulation speed compared to conventional simulation techniques”, explains Salmela.
Another important aspect of Salmela’s dissertation is overcoming some of the experimental difficulties in ultrafast optics. Direct time-domain measurements of ultrafast instabilities in optics have remained a big challenge for researchers in the field due to extremely short timescales of 10’s of femtoseconds. In his dissertation, Salmela demonstrates that neural networks can accurately correlate the temporal properties of light with the spectrum that is significantly easier to obtain.
“The methods introduced in the dissertation are very general and can be of interest in other fields of nonlinear science as well”, Salmela concludes.
Lauri Salmela is originally from Sahalahti (Kangasala) and currently works at Huawei Tampere Camera Lab.
Public defence on Friday 1 December
The doctoral dissertation of MSc (Tech) Lauri Salmela in the field of physics titled Application of Machine Learning to Nonlinear Dynamics in Optical Fibers will be publicly examined at the Faculty of Engineering and Natural Sciences at Tampere University at 13 o’clock on Friday 1 December 2023 at Hervanta Campus, Tietotalo building in TB104 auditorium (Korkeakoulunkatu 1, Tampere). The Opponent will be Professor Sylvain Gigan from the Sorbonne Université, France. The Custos will be Professor Goëry Genty from the Faculty of Engineering and Natural Sciences, Tampere University.
Photograph: Ekaterina Krutova