PHYS 503 :: Physics Illinois :: University of Illinois at Urbana-Champaign
Instrumentation Physics: Applications of Machine Learning
Topics Covered
Unit 1: Data Science (3 weeks)
- Introduction to data science and machine learning in the physical sciences and related fields
- Scientific python environment
- Notebooks and numerical python
- Handing, visualizing and finding structure in data
- Dimensionality and Linearity
- Adapting linear methods to nonlinear problems
- Kernel functions
- Jet recognition and clustering algorithms at high energy colliders
- Visualization of complex data (and data simulations) from the Large Hadron Collider, the Deep Underground Neutrino Experiment, the Laser Interferometer Gravitational-Wave Observatory, and the Large Synoptic Survey Telescope (LSST)
- Collaborative analysis tools—scientific python, notebooks, and JupyterLab — used by LSST
- Modeling long time-scale beam instabilities in high energy storage rings
Example physics connections and investigations:
Unit 2: Probability and Statistics (1 week)
- Probability Theory
- Probability Density Estimation from data
- Statistical methods
- Framing the concepts: determining the parameters of fundamental physics from data sets with complex backgrounds
Example physics connections and investigations:
Unit 3: Bayesian Inference (3 weeks)
- Introduction to Bayesian statistics
- Stochastic Processes, Markov Chains and Markov Chain Monte Carlo
- Variational Inference
- Optimization
- Computational Graphs and Probabilistic Programming
- Bayesian Model selection
- Learning in a Probabilistic context
- The physics of cardiac neurology: predictable onset of arrhythmias in nonlinear nerve networks
- Framing the concepts: Occam's razor and Bayesian inference in a comparison of Ptolemaic epicycles with a Newtonian heliocentric model
Example physics connections and investigations:
Unit 4: Supervised Learning (1 week)
- Supervised Learning in Scikit-Learn
- Cross Validation
- Classification of galaxies based on Dark Energy Survey images
- Bounding the phase space of electromechanical systems in precision atomic physics experiments
Example physics connections and possible investigations:
Unit 5: Learning and Inference using Artificial Neural Networks (ANNs) (1 week)
- Introduction to ANNs
- Types of Neural Networks
- Loss functions
- Backpropogation and Training
- Suppression of quantum chromodynamics backgrounds in rare-process searches
- Incorporating known symmetries of fundamental physics into loss functions
Example physics connections and possible investigations:
Unit 6: Deep Learning (4 weeks)
- Convolutional Neural Networks
- Unsupervised learning networks
- Autoencoder networks
- Recurrent Networks
- Graph Networks and Graph Neural Networks
- Deep Reinforcement Learning
- Deep Neural Networks for classification and regression analysis of diagnostic medical imagery
- Convolutional Neural Network approach to casting particle physics detector data as an image classification problem
- Time-domain anomaly detection in sky surveys using the Open Supernova Catalog
- Fast simulation of calorimeter response in particle physics detectors using variational autoencoder networks
- Using graph neural networks for charged particle tracking in neutrino and collider experiments
- Neural message passing and interaction networks for the quantum properties of organic molecules
- Diagnosis of congenital cardio-pulmonary pathologies through field analysis of acoustic, myographic, and electrical anomalies in infants
- Applications of reinforcement learning in controlling particle beams and confined plasmas
Example physics connections and possible investigations:
Unit 7: Methods for accelerated machine learning and inference (1 week)
- GPU accelerators;
- Distributed learning;
- Role hardware accelerators in ML inference
- The physics and information science issues of multi-spectral system configuration: time-alignment of data from distributed sensor arrays linked by affordable but unreliable networks
- Application of fast ML inference in multi-messenger astrophysics (e.g. supernovae detection)
Example physics connections and possible investigations: