University of Leeds Featured PhD Programmes
Catalysis Hub Featured PhD Programmes
University of Kent Featured PhD Programmes
University of Liverpool Featured PhD Programmes
University of Reading Featured PhD Programmes

Adversarially-robust neural networks for cyber security


Project Description

Deep neural networks (DNNs) have proven to be an extremely effective technology, most notably in image classification and speech recognition. They have also been proposed for a variety of tasks related to security, such as malware classification, classification of user content, and intrusion detection. However, there are significant challenges in using any kind of machine learning in security. Examples include:
• The high cost of classification errors
• The difficulty explaining outcomes to users (e.g., why a particular action they are attempting is being rejected)
• The requirement for constant evolution, as adversaries adapt to work around defences based on machine learning.
• The likelihood of weaponisation of user feedback. If users are given explanations why their actions are being denied, this can be used by attackers to devise ways to defeat the defences.

A related issue is that, in security, we often want to detect anomalies; but DNN techniques work by finding similarities, between the training data and new test data, not by finding outliers. Hence, the security context looks mismatched with the capabilities of DNNs. The objectives of the project are:
1. To characterise situations in which DNNs can, and cannot, be effectively used in security contexts.
2. To create techniques for making DNNs more robust against adversaries.
3. To explore the effectiveness of model transferability and black-box attacks.

Candidates must be UK nationals willing to undergo security clearance and will undertake a short internship with GCHQ. Candidates should have an honours undergraduate and/or postgraduate degree with Distinction (or an international equivalent) in Electrical and Electronics Engineering, Computer Science, Mathematical Engineering or closely related discipline. Familiarity with machine learning and neural nets and knowledge of cyber security is advantageous but candidates with a strong academic record will also be considered.

Funding Notes

£22,000 per year for 3.5 years, rising to £23,000 in the third year and EU/UK tuition fees, laptop, equipment, software and travel to attend conferences and summer schools.

Related Subjects

How good is research at University of Birmingham in Computer Science and Informatics?

FTE Category A staff submitted: 40.60

Research output data provided by the Research Excellence Framework (REF)

Click here to see the results for all UK universities

Email Now

Insert previous message below for editing? 
You haven’t included a message. Providing a specific message means universities will take your enquiry more seriously and helps them provide the information you need.
Why not add a message here
* required field
Send a copy to me for my own records.

Your enquiry has been emailed successfully





FindAPhD. Copyright 2005-2019
All rights reserved.