AI based vision recognition and decision making systems are being employed in a number of application domains including cyber physical systems, medical imaging, security, and autonomous systems. The development and application of deep learning approaches in these systems have in recent years become very sophisticated, partly due to advances in deep leaning techniques, and partly due to advances in hardware acceleration and support. However their employment in safety-critical and high reliability applications is restricted for two reasons. Firstly, traditional Software Engineering verification and validation (V&V) approaches do not readily extend to the black-box nature of these systems. Secondly, there are no guarantees of false positives or negatives, and attempts to correct, or to continue to learn in a deployed system, break all classical V&V processes.
UK industrial strategy promises the deployment of autonomous vehicles in some capacity on UK roads by 2021. However the sector is still understanding how safety and reliability of autonomous vehicles will be regulated and enforced, and what V&V technologies are needed to support this.
The significant challenges facing autonomous vehicle V&V can be summarised in two distinct points. Firstly, no precise specification for safety exists – while reduced number of road deaths is a primary goal, it is not in itself a formal specification. Secondly, variation in operating environments is unconstrainable. For instance subtle changes in circumstance that are easily human detectable such as a small child meandering near a crossing can lead to massively different scenarios to a vision recognition and decision making systems.
This project will investigate the sensitivity of a deep learning vision recognition system to subtle changes in test case scenario presentation. The goal of the project is two fold: Firstly, to objectively quantify the differences in scenarios (presented in an open source format) that impact on overall safety case analysis. Secondly to develop an insight into the limits of scenario based (simulation) testing and validation for safety-critical applications and objectively demonstrate these limitations.
Applicants are required to hold/or expect to obtain a UK Bachelor Degree 2:1 or better in a relevant subject. The University of Leicester English language requirements apply where applicable.
How to apply
The online application and supporting documents are due by Monday 21st January 2019.
Any applications submitted after the deadline will not be accepted for the studentship scheme.
References should arrive no later than Monday 28th January 2019.
Applicants are advised to apply well in advance of the deadline, so that we can let you know if anything is missing from your application.
1. Online application form
2. Two academic references
4. Degree certificate/s (if awarded)
5. Curriculum Vitae
6. CSE Studentship Form
7. English language qualification
Applications which are not complete by the deadline will not be considered for the studentship scheme. It is the responsibility of the applicant to ensure the application form and documents are received by the relevant deadlines.
All applications must be submitted online, along with the supporting documents as per the instructions on the website.
Please ensure that all email addresses, for yourself and your referees, are correct on the application form.
Project / Funding Enquiries
Application enquiries to [email protected]
Closing date for applications – 21st January 2019