- Student Records
Programme & Unit Catalogues

MA50263: Mathematics of machine learning

Follow this link for further information on academic years Academic Year: 2019/0
Further information on owning departmentsOwning Department/School: Department of Mathematical Sciences
Further information on credits Credits: 6      [equivalent to 12 CATS credits]
Further information on notional study hours Notional Study Hours: 120
Further information on unit levels Level: Masters UG & PG (FHEQ level 7)
Further information on teaching periods Period:
Semester 1
Further information on unit assessment Assessment Summary: CW 100%
Further information on unit assessment Assessment Detail:
  • Computational Project (CW 50%)
  • Research Project (CW 50%)
Further information on supplementary assessment Supplementary Assessment:
Like-for-like reassessment (where allowed by programme regulations)
Further information on requisites Requisites: Must have Programming ability in Python or other high-level language. Graduate level mathematics skills.
In taking this module you cannot ( take CM50264 OR take CM50265 )
Further information on descriptions Description: Aims:
To teach Machine Learning, including theoretical background and tools for implementation, to statistical applied mathematicians.

Learning Outcomes:
After taking this unit, students should be able to:
* Demonstrate knowledge of modern machine learning techniques
* Use computational tools for applying machine learning
* Show awareness of the applications of these methods
* Understand the mathematical models underlying machine learning algorithms and details of their implementation
* Write the relevant mathematical arguments in a precise and lucid fashion.

Problem Solving (T,F&A), Computing (T,F&A), independent study and report writing.

Introduction to machine learning (supervised vs unsupervised learning, generative vs discriminative models, validation, regression vs neural networks, computational tools in Python).
Additional topics will be chosen from:
* Neural networks (feed-forward, convolution, recurrent networks). Universal approximation theorem. Gradient descent
* Graphical models (decision trees, random forests, Markov random fields, Boltzmann machines)
* Bayesian non-parametric (Gaussian and Dirichlet process regression, hyper parameters)
* Reinforcement learning
* Shrinkage methods.
Further information on programme availabilityProgramme availability:

MA50263 is Optional on the following programmes:

Department of Mathematical Sciences