PhD studentship in the area of Countering Adversarial Attacks in Deep Reinforcement Learning Agents.
- Qualification Type: PhD
- Hours: Full Time
- Title of project: Countering Adversarial Attacks in Deep Reinforcement Learning Agents
- Closes: 24th June 2022, or until places have been filled.
Autonomy is increasing rapidly as machine learning techniques such as Deep Reinforcement Learning become more prevalent in the guidance of autonomous platforms.
As the use of AI has increased researchers have found that those intelligent guidance/control schemes are vulnerable to certain types of attack.
These adversarial attacks make imperceptible changes to the inputs to alter the decisions of the AI agent. These attacks work by making specific changes to the transformations that guide the AIs to the correct decisions.
Most research on countering adversarial attacks has been on classifiers which is why this project will focus on the less explored impact on DRL agents.
DRL has become a popular method recently and has been used in applications such as autonomous driving and systems control.
It has tended to be implemented in applications with a need for high safety standards as its use of continuous action spaces allows it to be more reactive to a wider range of situations.
Developing ways to deal with adversarial attacks is a necessary step to the wide adoption of some of these autonomous technologies.
On this project and in conjunction with a prestigious research organisation we will look to develop novel solutions to the potential attacks on DRL agents.
A successful applicant will be responsible for conducting supervised and independent research and all that entails, working with the supervisor, collaborating with the rest of the research team, and the research partner and producing papers for publishing in journals and for conferences.
The candidate will be based at City University in London for the duration of the PhD.
This project aims to develop novel solutions to the potential attacks on DRL agent and will focus on the less explored impact on DRL agents.
Eligibility and requirements
Suitable applicants must be solely British and hold a First-Class UK Master’s degree in electrical, mechanical, or aeronautical engineering.
They should have experience in machine learning and Python programming and particularly experience with the common machine learning packages.
They should have previously conducted some research in this or a related field.
The studentship is for 3 years and will provide full coverage of tuition fees (Home) and an annual tax-free stipend of £12,000 minimum.
How to apply
If you are interested in applying, you are encouraged to email initial informal enquiries to Nabil Aouf.
Visit our Electrical and Electronic Engineering research degrees web page for further information on making a formal application.
When submitting your application, enter the title “Countering Adversarial Attacks in Deep Reinforcement Learning Agents” and you will automatically be considered for this studentship.
You do not need to submit a proposal as part of your application as the project has already been outlined.
The online application can be found in the ‘How to apply section’ in the web link above and should include the following supporting documents:
- Copies of Degree Certificates and Transcripts in official English translation - original will be requested before an offer is made.
- Official work e-mail addresses (not private ones) for two referees (one of which must be an academic).
- Proof of English Language proficiency (minimum average score of 6.5 IELTS, with a minimum of 6.0 in each of the four components) if English is not your first language.
The outcome of the selection process should be announced by the end of July. The successful candidate will formally start their doctorate in October 2022.
For queries regarding the application process, please email the School.
Equality, diversity and inclusion
City, University of London is committed to promoting equality, diversity and inclusion in all its activities, processes, and culture, for our whole community, including staff, students and visitors.
We welcome applications regardless of gender, sexual orientation, disability, marital status, race, nationality, ethnic origin, religion or social class. For more information on our approaches to encouraging an inclusive environment, please see our Equality, Diversity and Inclusion pages.