Towards ML-Based Diagnostics of Laser-Plasma Interactions.

Yury Rodimkov,Valentin Volokitin,Evgeny Efimenko,Alexey Polovinkin,Iosif Meyerov,Mattias Marklund,Arkady Gonoskov,Thomas Blackburn,Shikha Bhadoria

doi:10.3390/s21216982

Abstract

The power of machine learning (ML) in feature identification can be harnessed for determining quantities in experiments that are difficult to measure directly. However, if an ML model is trained on simulated data, rather than experimental results, the differences between the two can pose an obstacle to reliable data extraction. Here we report on the development of ML-based diagnostics for experiments on high-intensity laser–matter interactions. With the intention to accentuate robust, physics-governed features, the presence of which is tolerant to such differences, we test the application of principal component analysis, data augmentation and training with data that has superimposed noise of gradually increasing amplitude. Using synthetic data of simulated experiments, we identify that the approach based on the noise of increasing amplitude yields the most accurate ML models and thus is likely to be useful in similar projects on ML-based diagnostics.

Highlights

Machine learning (ML) methods open novel ways for solving long-standing problems in many areas of physics, including plasma physics [1,2,3,4], condensed-matter physics [5,6], quantum physics [7,8,9,10], thermodynamics [11], quantum chemistry [12], particle physics [13]and many others [14,15]
We report on the developments of the ML-based diagnostics for the experiments on high-intensity laser–target interactions, in the area where the difficulty of experimental diagnostics is known to hamper a broad range of theoretically anticipated breakthroughs, ranging from unique XUV sources [18,19] to compact particle accelerators [20,21]
We trained the neural network (NN) 3 times to find the optimal number of principal components, and 10 times in other experiments

Summary

Introduction

Machine learning (ML) methods open novel ways for solving long-standing problems in many areas of physics, including plasma physics [1,2,3,4], condensed-matter physics [5,6], quantum physics [7,8,9,10], thermodynamics [11], quantum chemistry [12], particle physics [13]and many others [14,15]. One prominent problem for ML methods is to generalize results of numerical simulations, such that they can be used to extract unknown information from experimentally measured data. In contrast to conventional approaches, for which a particular feature in the data is exploited, ML methods are designed to learn indicative features when trained to solve the inverse problem, i.e., determine the input of simulation based on the outcome. A particular challenge is how to prevent the ML model from learning and making decisions based on features that are an artifact of synthetic data (results of simulation) only, i.e., on the features that are not present in the experimental data. One general approach is based on adding noise (jitter) to the input data or on varying parameters of extended parameter space [16,17]. We report on the developments of the ML-based diagnostics for the experiments on high-intensity laser–target interactions, in the area where the difficulty of experimental diagnostics is known to hamper a broad range of theoretically anticipated breakthroughs, ranging from unique XUV sources [18,19] to compact particle accelerators [20,21]

Methods

Results

Conclusion