Throughout industry and science, optimisation tasks require the trading-off of multiple quality criteria which are in competition with one another. This multi-objective optimisation task often requires the search and return of a set of solutions rather than a single design. This module spans specialised ‘expensive’ optimisation approaches where the cost function can only be queries a few hundred times at most, through to those designed for real-time optimisation of multi-objective optimisation problems which change over time, and robust optimisation. It also covers multi-criterion decision making – which is concerned with how a final design is selected from a trade-off set. 

For MSc students who will not have taken ECM2423, it is recommended to take ECMM409 (but it is not a requirement).

The aim of this module is to give you a theoretical and practical understanding of the optimisation of black-box multi-objective problems. By the end of this module you should be able to recognise the different sub-tasks and problems within a multi-objective optimisation task, and reason through the appropriate selection of an optimiser. You should also be able to directly undertake decision making process for solution selection, or guide a problem owner through this. You should be able to use different multi-objective optimisation algorithms implemented in different libraries.

Module Specific Skills and Knowledge
1 Demonstrate a clear understanding of the main categories of multi-objective optimisation;
2 Demonstrate a clear understanding of a range of multi-objective decision-making processes;
3 Implement a multi-objective optimisation software pipeline, and evaluate its performance;

Discipline Specific Skills and Knowledge:
4 Demonstrate familiarity with the main trends in multi-objective optimisation research;
5 Choose and use an appropriate development process
6 Implement software for addressing real-world optimisation problems;

Personal and Key Transferable/ Employment Skills and Knowledge:
7 Read and digest research papers from conferences and journals;
8 Relate theoretical knowledge to practical concerns;
9 Conduct a research project including sound statistical analysis of experimental results, and contrast the results found with those expected given previously published material;
10 Tackle a significant technical problem, and communicate the results.

The module content will be delivered by a mixture of lectures, workshops and directed reading. Indicative topics to be covered in the module include:

The multi-objective optimisation task;

Quality measures in multi-objective optimisation;

Multi-objective optimisation of expensive problems; 

Machine learning and Multi-objective optimisation;

Multi-objective optimisation of noisy problems;

Multi-objective optimisation of robust problems;

Multi-objective optimisation of dynamic problems;

Computational efficiency considerations in multi-objective optimisation;

Many- versus multi-objective optimisation;

Multi-objective cost/fitness landscapes;

Visualising multi-objective problems;

Preference incorporation in multi-objective optimisation;

Interactive multi-objective optimisation;

Hybridisation of evolutionary and multiple-criteria decision making.

Reassessment will be by coursework and/or written exam in the failed or deferred element only. For referred candidates, the module mark will be capped at 50%. For deferred candidates, the module mark will be uncapped.

Web based and electronic resources:

• Desdeo framework: https://desdeo.misitano.xyz/about/
• Python libraries: 
  • https://deap.readthedocs.io/en/master/
  • https://pymoo.org/operators/index.html
  • https://desdeo.misitano.xyz/
  • https://github.com/jMetal/jMetalPy
  • https://platypus.readthedocs.io/en/latest/#
  • https://esa.github.io/pygmo/
• Evolutionary multi-objective optimisation resources: https://neo.lcc.uma.es/emoo/

Reading list for this module: