MTH4331 - Optimisation for data analytics - 2019

This unit covers the theory, techniques and applications of optimisation, with a focus on applications in data analytics. The emphasis is on advanced methods for nonlinear continuous optimisation. In addition to its theoretical description of optimisation algorithms, the unit also has a strong practical focus with students required to solve problems computationally through programming. Topics covered include a selection from quasi-Newton methods, augmented Lagrangian methods, and stochastic gradient descent methods, with applications to machine learning and neural networks. Furthermore, the unit will cover constrained optimisation methods that may include quadratic programming, interior point methods, as well as stochastic meta-heuristics for nonlinear optimisation. Applications of these methods may include support vector machines and other classification methods.

On completion of this unit students will be able to:

1. Develop specialised mathematical knowledge in nonlinear optimisation algorithms and their efficient computer implementation
2. Understand the connection between optimisation and the training of data science models.
3. Determine an appropriate choice of optimisation approach based on problem characteristics.
4. Apply sophisticated optimisation methods to large problems arising from data analytics
5. Translate the result of optimisation into the application domain
6. Apply critical thinking in the field of computational optimisation

NOTE: From 1 July 2019, the duration of all exams is changing to combine reading and writing time. The new exam duration for this unit is 3 hours and 10 minutes.

Examination (3 hours): 60% (Hurdle)

Continuous assessment: 40%

Hurdle requirement: To pass this unit a student must achieve at least 50% overall and at least 40% for the end-of-semester exam.

• 3 hours of lectures and 1 hour tutorial per week
• 8 hours of independent study per week

See also Unit timetable information

Master of Mathematics