Evolutionary Robotics
The PhD project will explore optimisation, learning and/or adaptation in the context of evolutionary robotics. Possible avenues of research include the co-evolution of morphology and control of robots the interaction of evolution and learning mechanisms to produce bodies and behaviours that are specialised to specific environments and tasks. Alternative projects might focus on adaptation of behaviour only, using learning methods (e.g. evolution, reinforcement learning) to adapt controllers in real time to adapt to new environments, or learning repertoires of behaviours to enable robust performance. Another promising area is in the use of state-of-the art methods from the quality-diversity literature to fully explore rich search spaces of both morphologies and controller. Projects can be conducted in simulation only but there is also the possibility to utilise our robotics laboratory to conduct experiments on physical robots.
Continual Learning in Combinatorial Optimisation
Combinatorial problems are ubiquitous across many sectors. In a typical scenario, instances arrive in a continual stream and a solution needs to be quickly produced. Meta-heuristic search techniques have proved useful in providing high-quality solutions, but it is challenging to select the correct solver for a particular instance and/or tune it to optimise performance. If the characteristics of instances change over time, it is also possible that at some future point, instances are sufficiently novel that there is no appropriate solver known or the selector is incapable of choosing the best algorithm. This project will focus on one or more aspects of tackling this issue; for instance developing novel algorithm-selection methods that are capable of selecting the most appropriate method; using algorithm-generation methods (e.g. genetic programming) to generate or tune algorithms to work well on instances that occur in novel regions of the instance space; developing methods that are capable of learning from experience, i.e. continually improving selection methods or generation methods over time as knowledge is learned from solving past instances. The project is likely to mix techniques from meta-heuristic optimisation, automated algorithm generation and machine-learning, particularly borrowing ideas from the transfer learning or continual learning literature.
Academic Qualifications
A first degree ideally in Computer Science, however a degree in another scientific subject with a good fundamental knowledge of computer science is also acceptable.
English language requirement
IELTS score must be at least 6.5 (with not less than 6.0 in each of the four components). Other, equivalent qualifications will be accepted. Full details of the University's policy are available online.
Essential attributes:
- Good knowledge of the fundamental concepts of Stochastic Search Methods including Evolutionary Computation
- Highly competent programming skills (e.g. in Python/C++), and good knowledge of statistics and data-analysis
- Knowledge of machine-learning methods (particularly for Project 2)
- Good written and oral communication skills
- Strong motivation, with evidence of independent research skills relevant to the project
- Good time management
Continue with Facebook
