Trusted Learning Systems for Cyber-Security Applications

Project Title:

Trusted Learning Systems for Cyber-Security Applications


Joarder Kamruzzaman

Contact person and email address:

A brief description of the project:

Intelligent systems based on machine learning (ML) techniques constitute the core of many everyday applications today, and in particular computer security paradigm is largely built around machine learning algorithms. However, adversarial techniques in recent time have become more sophisticated to fool ML techniques, which may trigger unwanted consequences in our effort to create highly interactive and smart living environments. For example, a vision recognition system of a driverless car can be fooled to misrecognise a ‘stop’ road sign with minimum alteration to the sign or with minor modification a hacker might be able to evade a malware detection system, if the ML algorithms running in those systems are not robust enough. All these offer significant challenges to design ML system which can be regarded trustworthy. This project will investigate ways to assess robustness of ML algorithms and explore techniques to increase their resilience, with particular applications in cyber security domain.