Date: Tuesday 2 August 2022 – 17:15 (Europe/London)
Speaker: Dr Adi Hanuka, Senior Software Engineer, Machine Learning, Eikon Therapeutics, CA, US
Abstract
Phase space measurement is one of the key diagnostics in particle accelerator machines. Existing beam diagnostics are invasive, and oftentimes cannot operate at the required resolution. In this work we present the Virtual Diagnostic (VD) tool, a computational tool based on deep learning that can be used to predict a diagnostic output. We show how VD accurately predicted the electron beam longitudinal phase space (LPS) for every shot using spectral information collected non-destructively from the radiation of a relativistic electron beam. In addition, we show both experimental and simulated examples where VD helps overcome resolution limitations (e.g. high repetition-rate machine or high-current ultra-short bunch).
We then show how we quantified the uncertainty of the VD prediction, its robustness against out-of-distribution inputs, and the information extracted from the latent space representations.
VDs are especially useful in systems where measuring the output is invasive, limited, costly or runs the risk of altering the output. They can provide confident knowledge while reducing the load on data storage, readout and streaming requirements. In combination with quantified uncertainties, VDs can enable making informed decisions for safety-critical systems such as particle accelerators.
Refs:
- Accurate and confident prediction of electron beam longitudinal properties using spectral virtual diagnostics
- Virtual diagnostic suite for electron beam prediction and control at facet-ii
- Uncertainty quantification for virtual diagnostic of particle accelerators
- Aberration Corrector Tuning with Machine-Learning-Based EmittanceMeasurements and Bayesian Optimization
- Uncertainty quantification for deep learning in particle accelerator applications
Biography
Dr Hanuka is an electrical engineer, particle accelerator physicist, and AI researcher. She works at the intersection of physics and machine learning (ML), focusing on how ML can build better systems, and how physics can help build better ML algorithms.
Adi earned her PhD in Electrical Engineering from the Israel Institute of Technology - in the field of optical particle accelerators, during which time she collaborated with the Fermilab, UCLA and SLAC as part of the “Accelerator on Chip” international program. In 2019, Adi joined SLAC to continue developing the field of ML for particle accelerators. Adi joined Eikon Therapeutics in mid-2021 with the aim of using artificial intelligence to develop new drugs.
Hanuka was named to the Forbes’ Israel “30 under 30” list of promising young scientists, and her scientific contributions were acknowledged by numerous awards including the Schmidt Foundation Award and the Rothschild Fellowship. Adi is the editor of the “Operation Intelligence” research topic in the ``Frontiers Big Data and AI in High Energy Physics" journal. More recently, Adi taught the US Particle Accelerator School (USPAS) course: “Machine Learning and Optimization for Particle Accelerators”.
You can now watch the seminar on YouTube: https://youtu.be/panTtEij6rQ
