Fairness in Markov decision processes
Fairness is a desirable property of decision rules applied to a population of individuals. For example, college admissions should be decided on variables that inform about merit, but fairness may also require taking into account the fact that certain communities are inherently disadvantaged. At the same time, a person should not feel that another in a similar situation obtained an unfair advantage. All this must be taken into account while still caring about optimizing for a decision maker's utility function. As another example, consider mortgage decisions: while lenders should take into account the creditworthiness of individuals in order to make a profit, society must ensure that they do not unduly discriminate against socially vulnerable groups. The problem becomes even more challenging when we take into account potential uncertainties in decision making models, which can make some notions of fairness impossible to satisfy. This project will examine fairness in decision making for a topic of the student's choice, but the focus should always be in data driven problems.
The project can be either theoretical or practical.
For the former, a solid background in mathematics is necessary,
as the student will attempt to develop and prove fairness
properties of algorithms. For the latter, the student should be
able to envisage an appropriate case study, such as mortgage
decisions or university admissions, collect data for it.
Mathematical maturity is still necessary, but the emphasis is
more on research methodology and programming skills. For this
particular project, our emphasis is on the sequential aspects of
fairness, more specifically fairness in Markov decision
processes. This includes the tasks:
- Formalize the decision problem.
- Specify the fairness criteria used.
- Develop an algorithm solving the problem while satisfying the fairness criteria.
Skills. The student should have good working knowledge of
calculus, probability and algorithms.
Benefits. The student will obtain background in
statistics, machine learning and constrained optimisation, as
well the links between these areas, as well as practical
experience in programming (mainly python/octave with some
performance critical functions in C) and/or optimality proofs.
Good theoretical or experimental results will lead to writing and
submitting a paper to a suitable peer-reviewed