ES98E-15 Scientific Machine Learning
This module provides students with knowledge in the modern field of scientific machine learning, which is a fusion of scientific computing and machine learning.
Understand how to use a variety of statistical and machine learning techniques to train models which combine data-driven and mechanics models and assess their ability to make useful predictions.
This is an indicative module outline only to give an indication of the sort of topics that may be covered. Actual sessions held may differ.
- Bayesian linear regression and enhancements to it such as automatic relevance determination
- Gaussian process regression and enhancements such as sparse GP regression and the Kennedy-O'Hagan approach to model discrepancy.
- Kernel based ML-techniques including SVM and applications of reproducing kernel Hilbert spaces
- Dimensionality reduction as a tool for data analysis (PCA, kPCA, active subspaces)
- Advanced state space models such as the ensemble Kalman filter
- Neural networks and deep neural networks
- Neural differential equations and physics informed neural networks
- Advanced approximate Bayesian inference: variational inference, expectation maximisation, etc.
By the end of the module, students should be able to:
- Understand and interpret the application of Bayesian inference to infer model parameters from data
- Apply Gaussian process regression to build surrogate models for computational codes and evaluate the results
- Apply deep neural networks to accelerate scientific computing and interpret the results
- Synthesise neural network and mechanistic models and apply them to perform scientific machine learning
- Evaluate advanced inference techniques such as variational inference work and critique when they can be applied
- Recognise, formulate, analyse and interpret machine learning solutions to scientific problems
- Critique the application of machine learning methods to cutting edge problems in scientific computing
Indicative reading list
Bishop, Pattern Recognition and Machine Learning
MacKay, Information Theory, Pattern Recognition and Learning Algorithms
Rasmussen and Willians, Gaussian Processes for Machine Learning
Gelman, Bayesian Data Analysis
Hastie, The Elements of Statistical Learning: Data Mining, Inference and Prediction
The MSc programme will recruit students with backgrounds across the physical and mathematical sciences, including engineering, and will provide an interdisciplinary perspective on predictive modelling.
Scientific machine learning is a fusion of scientific computing and machine learning, drawing from mathematics, statistics and the modelling of physical phenomena across a wide range of application domains.
Subject specific skills
- Machine learning
- Computational statistics
- Predictive modelling
- Fusion of advanced data analysis and mathematical modelling techniques
- Data analysis and modelling
- Oral presentation skills
- Scientific computing
|Lectures||10 sessions of 2 hours (13%)|
|Supervised practical classes||8 sessions of 2 hours (11%)|
|Private study||114 hours (76%)|
Private study description
Students will work independently to complete the weekly assignments outside of the workshops and to prepare for the viva.
No further costs have been identified for this module.
You must pass all assessment components to pass the module.
Assessment group D
|Computer Laboratory Assignments||60%|
6 x 1 page assignments based upon lecture topic and computer laboratory work
Feedback on assessment
Annotation of computer workbooks with feedback on individual questions
Written feedback from examiners of viva voce exam
This module is Core for:
- Year 1 of TESA-H1B1 Postgraduate Taught Predictive Modelling and Scientific Computing