When swarm-evolutionary optimization outperforms other optimization algorithms for data-analysis.

What is it about?

The optimization of the free parameters of a theoretical model to fit experimental data often is a cumbersome process. In the recent years, one has witnessed a transition from building on the experience of a human scientist to the almost blind trust in the capabilities of so-called artificial intelligence, mostly relying on artifical neural networks (ANN). The present work illustrates that there exists an extensive list of optimization algorithms, and the choice which one to use should be based on a good understanding of the problem and on a thorough investigation of the capabilities of the various algorithms. In the present publication, we show that the swarm-based meta-heuristic optimization of differential evolution disruptively will improve the precision and accuracy in materials characterization. The algorithm can find the optimal model parameters, even if they are highly correlated. Thereby, Rutherford backscattering spectrometry, which already has the reputation of being very accurate and reference-free, now also gains importance by being more precise.

Featured Image

Photo by James Wainscoat on Unsplash

Photo by James Wainscoat on Unsplash

Why is it important?

A new level of absolute accuracy and precision in Rutherford backscattering spectrometry is demonstrated. More generally, it highlights the importance of estimating and reporting uncertainties, which is often neglected because of deemed difficult to do. This paper illustrates how the stochastic approach not only helps to find the parameters, but also helps to estimate the uncertainties in a straightforward manner, without the need for deep analytical approaches, only benefitting from a simple computational method.

Perspectives