Anomaly detection in X-ray physics
Abstract/Contents
- Abstract
- Anomaly detection is an important task for complex systems (e.g., industrial facilities, manufacturing, large-scale science experiments), where failures in a sub-system can lead to low yield, faulty products, or even damage to components. These systems produce too many signals for the system operators to monitor in real time. In addition, many of these signals are only interpretable by subject matter experts with years of experience. As a result, changes in system performance can require time-intensive consultations with experts to identify the underlying problem. For this reason, despite the wealth of data that these complex systems produce, labeled anomalies are typically rare (or even nonexistent) and expensive to acquire. Since there are no labels, the training data will inevitably be contaminated by anomalies. In Chapter 2, we introduce a fully automated, unsupervised beam-based diagnostic method for RF station fault detection at SLAC's Linac Coherent Light Source (LCLS). Our key contribution is the combination of high fidelity beam position monitoring data with bandwidth limited (and noisy) radio-frequency (RF) station diagnostic data to achieve highly accurate, nearly real time fault identification. In Chapter 3, we introduce coincident learning for anomaly detection (CoAD) and an unsupervised metric \hat{F}_\beta, that assumes and exploits that anomalies are coincident in two different slices of the input data. We present theoretical properties of our metric and demonstrate its performance on four data sets. In Chapter 4, we focus on a broader task of detecting any anomalous beam behavior at LCLS and develop the unsupervised Resilient Variational Autoencoder (ResVAE) model. We illustrate its resilience to contaminated training data and apply it to identifying anomalies in accelerator status at LCLS . In Chapter 5, we switch gears and address a problem in neural network pruning. We introduce a novel channel pruning approach for convolutional neural networks, called Soft Masking for cost-constrained Channel Pruning (SMCP), that is particularly suitable when pruning a large fraction of the network's channels. The core of our approach relies on regularly rewiring the network sparsity, through soft masking of the network weights, to minimize the accuracy drop for large pruning fractions. Our method outperforms prior works on both the ImageNet classification and PASCAL VOC detection datasets.
Description
Type of resource | text |
---|---|
Form | electronic resource; remote; computer; online resource |
Extent | 1 online resource. |
Place | California |
Place | [Stanford, California] |
Publisher | [Stanford University] |
Copyright date | 2023; ©2023 |
Publication date | 2023; 2023 |
Issuance | monographic |
Language | English |
Creators/Contributors
Author | Humble, Ryan A |
---|---|
Degree supervisor | Darve, Eric |
Thesis advisor | Darve, Eric |
Thesis advisor | Ermon, Stefano |
Thesis advisor | Ratner,Daniel |
Degree committee member | Ermon, Stefano |
Degree committee member | Ratner,Daniel |
Associated with | Stanford University, School of Engineering |
Associated with | Stanford University, Institute for Computational and Mathematical Engineering |
Subjects
Genre | Theses |
---|---|
Genre | Text |
Bibliographic information
Statement of responsibility | Ryan Alexander Humble. |
---|---|
Note | Submitted to the Institute for Computational and Mathematical Engineering. |
Thesis | Thesis Ph.D. Stanford University 2023. |
Location | https://purl.stanford.edu/hz309hf4498 |
Access conditions
- Copyright
- © 2023 by Ryan A Humble
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...