- Prospective Applicants
-
Current PhD Researchers
- Careers and Opportunities
- Careers
- Go Global
- Science Shop
-
Supervisors
- Find a PhD
- PhD Manager
Blindly trusting the predictions made by a machine learning model can lead to disaster. A more cautious approach is to consider the uncertainty in a model’s predictions (the “predictive uncertainty”), before taking any action based on them. However, just as we should not blindly trust a model’s predictions, nor should we blindly trust a model’s predictive uncertainties either, otherwise it may provide the user with nothing but a false sense of security. Having a reliable predictive uncertainty is important in a growing number of applications. For example, electrical grid operators routinely make forecasts concerning the energy output from renewable sources, such as solar and wind. Reliably quantifying the uncertainty in these forecasts enables the renewable energies to be incorporated into the grid more intelligently. Recently, methods have been proposed which can “calibrate” the predictive uncertainty of a model to ensure it is neither over-confident (consistently under-estimating predictive uncertainty) nor under-confident (consistently over-estimating predictive uncertainty).
This PhD project, based in the AI Research Centre at Ulster University, will develop further algorithms in this important research area, optimized for a number of different high-impact applications across science and engineering that require reliable predictive uncertainty.
Research directions include:
Calibration with small amounts of data for Bayesian optimization: Bayesian optimization (where predictive uncertainty is crucial), necessarily involves small amounts of training data. Yet existing calibration methods typically involve large calibration sets, separate from the training data. How can calibration be effectively applied when labelled data is limited, as is the case in Bayesian optimization? Can we transfer uncertainties from a related problem for which we have a large amount of labelled data? How does the performance of Bayesian optimization depend on the accuracy of predictive uncertainty in Gaussian Process models?
Calibration of multi-output regression models: Some applications involve the prediction of multiple outputs from a single training set. For example, in computational chemistry, machine learning is used to predict multiple properties of individual atoms in molecules, for use in “force-fields”. The importance of considering the predictive uncertainty of force-fields in applications is gradually being realised. Can efficient methods be developed to calibrate the multiple predictive uncertainties in a multi-output model simulteanously? Are there trade-offs that must be overcome?
Explanations of uncertainty: Methods from Explainable AI (XAI) aim to explain the predictions made by machine learning models. In principle, XAI can also be applied to explain the uncertainty in predictions. These explanations may be important in their own right, but could they also be used to calibrate the uncertainties?
The exact research challenges to be addressed in this project can be tailored to the interests and experience of the PhD candidate.
Applicants should hold, or expect to obtain, a First or Upper Second Class Honours Degree in a subject relevant to the proposed area of study.
We may also consider applications from those who hold equivalent qualifications, for example, a Lower Second Class Honours Degree plus a Master’s Degree with Distinction.
In exceptional circumstances, the University may consider a portfolio of evidence from applicants who have appropriate professional experience which is equivalent to the learning outcomes of an Honours degree in lieu of academic qualifications.
If the University receives a large number of applicants for the project, the following desirable criteria may be applied to shortlist applicants for interview.
The University offers the following levels of support:
The following scholarship options are available to applicants worldwide:
These scholarships will cover full-time PhD tuition fees for three years (subject to satisfactory academic performance) and will provide a £900 per annum research training support grant (RTSG) to help support the PhD researcher.
Applicants who already hold a doctoral degree or who have been registered on a programme of research leading to the award of a doctoral degree on a full-time basis for more than one year (or part-time equivalent) are NOT eligible to apply for an award.
Please note: you will automatically be entered into the competition for the Full Award, unless you state otherwise in your application.
The scholarship will cover tuition fees at the Home rate and a maintenance allowance of £19,000 (tbc) per annum for three years (subject to satisfactory academic performance).
This scholarship also comes with £900 per annum for three years as a research training support grant (RTSG) allocation to help support the PhD researcher.
Due consideration should be given to financing your studies. Further information on cost of living
Kendall, A. (2018). Deep learning is not good enough, we need Bayesian Deep Learning for safe AI: Available at: https://alexgkendall.com/computer_vision/bayesian_deep_learning_for_safe_ai/
Zelikman, E. et al. (2020). Short-term solar irradiance forecasting using calibrated probabilistic models, arXiv:2010.04715.
Kuleshov, V. et al. (2018). “Accurate uncertainties for deep learning using calibrated regression,” in International Conference on Machine Learning. PMLR, 2018, pp. 2796–2804.
Zelikman et al. (2020). “CRUDE: Calibrating regression uncertainty distributions empirically,” arXiv preprint arXiv:2005.12496, 2020.
Hawe, G. I. and Sykulski, J.K. (2007). Considerations of accuracy and uncertainty with Kriging surrogate models in single-objective electromagnetic design optimization. IET Science, Measurement & Technology, 1 (1), 37-47.
Handley, C. M., Hawe, G. I., Kell, D. B., Popelier, P. (2009) Optimal construction of a fast and accurate polarisable water potential based on multipole moments trained by machine learning, Physical Chemistry Chemical Physics 11 6365-6376.
Peterson, A. A. et al. (2017). Addressing uncertainty in atomistic machine learning, Phys.Chem.Chem.Phys., 19 10978.
Submission deadline
Monday 7 February 2022
12:00AM
Interview Date
March 2022
Preferred student start date
mid September 2022
Telephone
Contact by phone
Email
Contact by email