Horizon Digital Economy Research Centre PhD Studentship - Technology Lab

Horizon Digital Economy Research Institute - Horizon Centre for Doctoral Training in My Life in Data

PhD Studentship in partnership with the Defence Science and Technology Laboratory (DSTL)

Our digital identities will define the interfaces to the future services that we will use for entertainment, wellbeing, government, transport, energy, retail and finance. They will be constructed from our location history, our personal data, and digital records that capture who, and where, we are and the histories of our digital, and increasingly our physical and social, interactions.

The Horizon Centre for Doctoral Training in My Life in Data will therefore train a community of PhD students with the transdisciplinary skills to address the challenge of digital identity and personal data for the 21st century.

We are recruiting two PhD students to carry out research in partnership with the DSTL with a focus on Deceiving the Machine. The general idea is to explore and analyse ways to fool machine-learning algorithms. We are currently considering two interlinked topics:

Topic 1: Limitations of current state-of-the-art machine learning.

The goal would be to develop a framework or set of metrics/methods that could formally quantify limitations of current machine learning techniques. Ideally these techniques would include deep learning methods. Points to investigate would be: where do they work well? Where don’t they? What is their sensitivity to changes in the input (i.e. how easily can they be fooled by physical or digital camouflage)? Are there methods of camouflage/transformation that are able to fool multiple certain machine learning algorithms but not others?

Topic 2: Accurate camouflage/adversarial attack detection.

The goal is the creation and development of machine learning algorithms impervious to camouflage and/or adversarial attacks. It can be seen as an adjunct to the previous topic, but instead of investigating current methods, the contribution would be in creating machine learning resistant to possible attacks. Detection could include physical camouflage (i.e. the use of particular physical objects to “confuse”) and digital camouflage (i.e. through image transformations, also known as adversarial example detection)

In both cases, we are also interested in looking for ways for algorithms to detect attack attempts. Preferred candidates are expected to have a background in computer vision and machine learning.

The students will benefit from:

• A fully-funded four-year PhD programme that integrates a leading-edge research project with research training in transdisciplinary skills.
• At least one internship with our partner.
• An enhanced stipend of £16,800 per annum as well as a personal laptop.

We have funding for UK/Home and EU students. Application forms are downloadable from http://cdt.horizon.ac.uk/apply/current-opportunities and should be returned by email with a detailed CV, transcript, references and a statement of research interests (personal statement) lynn.rees@nottingham.ac.uk. Please quote ref HORIZON2 DSTL, Closing date: 23 February 2018, Shortlisting date: 5 March 2018, Interview date: 12 March 2018.