Instead of changing the parameters manually, Saga Helgadóttir’s new method allows the machine to decide which ‘rules’ are applicable to the problem at hand. Through many iterations, the machine learns how to make accurate predictions from the input data. For this to work, very large volumes of training data with accurate values are needed.
"In my case, working with particles and bacteria, the images are usually spotty. You can easily simulate those kinds of images. So after training the neural network on large quantities of simulated data, we applied it to actual experimental data – and were actually surprised by how well it worked!"
In her two final years as a doctoral student, Saga Helgadóttir will continue to develop the method. Her research has already garnered a lot of interest in her new method just from the pre-print version of the article – which is now published in Optica, the flagship publication of The Optical Society of America.
Saga Helgadóttir has been invited by a research group at the Max-Planck Institute in Germany to give a presentation. Furthermore, she plans to develop a software to run her method, which would help branch use of it out into the medical field of research.
COMPAMED-tradefair.com; Source: University of Gothenburg