Dr. Amelia Elizabeth Pollard, PhD

I'm Amy, the resident machine learning researcher at the Accelerator Science and Technology Centre (ASTeC) working on applying machine learning to particle accelerator research.
I attained my PhD in Computer Science at the University of Manchester, in the Machine Learning & Optimisation group with my thesis Multitask Learning, Biased Competetion, and Inter-task Interference

The majority of my work concerns the CLARA facility at Daresbury Laboratory, Warrington.

You can contact me at amelia.pollard[at]stfc.ac.uk

Papers


Machine Learning Approach to Temporal Pulse Shaping for the Photoinjector Laser at CLARA (Poster: PDF) IPAC '22

A. E. Pollard, W. Okell, D. Dunning, E. Snedden

The temporal profile of the electron bunch is of critical importance in accelerator areas such as free-electron lasers and novel acceleration. In FELs, it strongly influences factors including efficiency and the profile of the photon pulse generated for user experiments, while in novel acceleration techniques it contributes to enhanced interaction of the witness beam with the driving electric field. Work is in progress at the CLARA facility at Daresbury Laboratory on temporal shaping of the ultraviolet photoinjector laser, using a fused-silica acousto-optic modulator. Generating a user-defined (programmable) time-domain target profile requires finding the corresponding spectral phase configuration of the shaper; this is a non-trivial problem for complex pulse shapes. Physically informed machine learning models have shown great promise in learning complex relationships in physical systems, and so we apply machine learning techniques here to learn the relationships between the spectral phase and the target temporal intensity profiles. Our machine learning model extends the range of available photoinjector laser pulse shapes by allowing users to achieve physically realisable configurations for arbitrary temporal pulse shapes.


Machine Learning for RF Breakdown at CLARA (Poster: PDF) ICALEPCS '21

A. E. Pollard, A. J. Gilfellon, D. J. Dunning

Maximising the accelerating gradient of RF structures is fundamental to improving accelerator facility performance and cost-effectiveness. Structures must be subjected to a conditioning process before operational use, in which the gradient is gradually increased up to the operating value. A limiting effect during this process is breakdown or vacuum arcing, which can cause damage that limits the ultimate operating gradient. Techniques to efficiently condition the cavities while minimising the number of breakdowns are therefore important. In this paper, machine learning techniques are applied to detect breakdown events in RF pulse traces by approaching the problem as anomaly detection, using a variational autoencoder. This process detects deviations from normal operation and classifies them with near perfect accuracy. Offline data from various sources has been used to develop the techniques, which we aim to test at the CLARA facility at Daresbury Laboratory. These techniques could then be applied generally.


Learning to Lase: Machine Learning Prediction of FEL Beam Properties (Poster: PDF) ICALEPCS '21

A.E. Pollard, D.J. Dunning, M. Maheshwari

Accurate prediction of longitudinal phase space and other properties of the electron beam are computationally expensive. In addition, some diagnostics are destructive in nature and/or cannot be readily accessed. Machine learning based virtual diagnostics can allow for the real-time generation of longitudinal phase space and other graphs, allowing for rapid parameter searches, and enabling operators to predict otherwise unavailable beam properties. We present a machine learning model for predicting a range of diagnostic screens along the accelerator beamline of a free-electron laser facility, conditional on linac and other parameters. Our model is a combination of a conditional variational autoencoder and a generative adversarial network, which generates high fidelity images that accurately match simulation data. Work to date is based on start-to-end simulation data, as a prototype for experimental applications.