Machine Learning (ML) algorithms are proving to be a powerful tool in the analysis of scientific data, capable of using large training datasets of previously collected data to identify features and predict the results of new measurements. ML encompasses a large number of technologies and tool sets that can be used to explore different types of scientific data, in this project we expect to make use of convolutional neural networks and vision transformers which excel when applied to imaging data. Many of these have been encompassed in accessible python packages such as scikit-learn, TensorFlow and PyTorch.
The aim of this project is to use these powerful new tools to develop algorithms that can reliably identity and classify 3 dimensional peak shapes on detectors from x-ray Bragg diffraction from crystals. Basic algorithms have been in use for many years but regularly fail when attempting to work with weak signals and high backgrounds, or those with complicated shapes, such as those from magnetic nanodomains. Examples of such patterns are attached, the first shows a split peak from orthorhombic domains and the second a weak resonant-magnetic peak outside of the region of interest. Developing more reliable peak identification software could be used to further automate beamline operations – for example, determining automatically the existence of a magnetic reflection (usually a very weak peak) would allow the automation of magnetic structure determination, increasing the types of experiment currently possible in this field.
Beamline I16 specialises in measuring incredibly subtle scattering phenomena in a wide variety of environments – from magnetic scattering or charge ordering at cryogenic temperatures, to diffuse scattering from thin films. As such, the back-catalogue of scan data from I16 contains a huge number of detector images of Bragg reflections with a wide variety of shapes and backgrounds.
In this project, machine vision techniques will be utilized to extract coarse labels from the back-catalogue of I16 scan data, which will then be used to train neural networks to reliably perform peak identification. The successful completion of this project will produce a general tool available to any scattering experiment and help the beamline develop new automated procedures for future experiments.
Applications are now closed for this project.
Please apply via our online application portal. The vacancy that you are applying for is the "Summer Placements 2023" listing, you will then have the opportunity to select up to three projects to apply for. This project's reference is 23009SP.
Interviews will be scheduled for 6, 7, 8, and 9 February 2023.
If you are disabled and would like to be considered under the Disability Confident Scheme, please let us know via the online application process.
Please note that this role does not meet the required skill level for a Skilled Worker visa and therefore we would be unable to sponsor individuals due to the current UK Home Office immigration rules. To be appointed to the role, candidates will need to have the right to work in the UK without sponsorship from us.
Diamond Light Source is the UK's national synchrotron science facility, located at the Harwell Science and Innovation Campus in Oxfordshire.
Copyright © 2022 Diamond Light Source
Diamond Light Source Ltd
Diamond House
Harwell Science & Innovation Campus
Didcot
Oxfordshire
OX11 0DE
Diamond Light Source® and the Diamond logo are registered trademarks of Diamond Light Source Ltd
Registered in England and Wales at Diamond House, Harwell Science and Innovation Campus, Didcot, Oxfordshire, OX11 0DE, United Kingdom. Company number: 4375679. VAT number: 287 461 957. Economic Operators Registration and Identification (EORI) number: GB287461957003.