EPSRC Fully funded PhD Scholarship (3.5 years) in Explainable Machine Learning

The Studentship is for UK and EU Applicants ONLY


Fully funded PhD Scholarship in Explainable Machine Learning


The successful candidate will be based in the Department of Computer Science, Durham University - ranked 6th for Computer Science in the UK The Complete University Guide 2019 (Durham University ranked 6th overall and 74th in QS World University Rankings 2019)

This studentship will start from October 2019 and the successful applicant will receive a fully funded scholarship for 3.5 years (subject to satisfactory progression).
The successful candidate will be encouraged to publish their work at major national and international avenues, the candidate can expect their research to have significant impact across industry and academia. The post offers an outstanding opportunity to gain a strong research track record in an exciting and fast-moving area of machine learning and deep learning.

The studentship includes:
- Domestic fees for UK and EU students.
- A stipend equivalent to the Research Councils UK National Minimum Doctoral Stipend (£15.000) plus £2000 additional research costs to cover publications/travel expenses…

The successful applicant will have the opportunity to deliver lab demonstrating in order to gain teaching experience during their PhD and top up their stipend.


Project description

Research Context:
We live in an unprecedented era of successful machine learning. However, for this success to continue and for machine learning models to be deployed in critical fields, they must be transparent, providing explanations of their decisions in a human understandable way. These explanations are important to ensure algorithmic fairness, identify potential bias [1], consistent performance, and safety against adversarial attacks. A model should explain why a clinical diagnosis was given, or a self-driving car was involved in an accident, etc. This is valid for every field with a regulatory body, e.g. the Financial Conduct Authority, or where human safety is of concern [2].

The Challenge:
Explaining the decision made by a deep neural network is very difficult due to the highly non-linear nature of the network. A complete explanation of the outcome should involve explaining not only the representation of the data within the model but also its processing and in a format that is meaningful to a domain expert. The problem is exasperated with noise and uncertainty in the data, the labels, and the models themselves [3].

Research Questions:
I) How to embed domain expertise in the explanation of machine learning model outcomes?
II) How to incorporate uncertainty in the model and explain decisions under noisy and missing data?
III) How to explain decisions in a zero-shot learning scenario with no data from a new class?
V) Can the explanation approach work with current black-box models?

The work will build on three pillars: I) Layer-wise Relevance Propagation (LRP) II) Bayesian modelling III) Causal Inference. LRP is a method to identify important input features by running a backward pass in the neural network [4].

Entry Requirements

For entry to the PhD you will be required to have achieved a 2:1 bachelor’s degree in an appropriate subject, from a recognised university (or equivalent).
EU students would also require a minimum overall IELTS 6.5 score of which no element of less than 6.0 (or equivalent)
Further information can be found at https://www.dur.ac.uk/computer.science/postgraduate/research/

How to apply

• Apply for Degree of Doctor of Philosophy in Computing Science https://www.dur.ac.uk/study/pg/apply/
• State name of the lead supervisor as the Name of Proposed Supervisor
• State ‘EPSRC’ as Intended Source of Funding
• State the exact project title on the application form

When applying please ensure all required documents are attached:
- Your CV.
- Transcript for your most recent qualification.
- Copy of the certificate of the most recent qualification.
- Two reference letters.

For informal enquiries please contact Dr Al Moubayed ([Email Address Removed]) with a copy of your curriculum vitae.