Defence Security

Defence and security

Detecting Threats in Temporal Networks

Organised crime in the modern world is characterised by large networks of criminals operating across national borders and transacting via multiple communication channels. These networks are both evolving temporally (as the relationship between individuals changes) and are multi-layered (for example, communications, goods, money), making analysis complex. In addition these networks are characterised by a sparse structure, often with key players only loosely connected to the main network in order to reduce the likelihood of detection. Patterns of communication within the network over time can expose the different types of activity by members. The aim of the project is to better understand how and when changes in patterns of communication within the network can be indicative of imminent major criminal operations.

OPTIMA CIs: Professor Peter Stuckey

OPTIMA AIs: Dr John Betts

Industry Supervisors: Dr Campbell Wilson & Dr Janis Dalins

PhD Student: apply now

Assessing risk of database reconstruction attacks on aggregate tables and linear systems

The increasingly available personal information online and the rise of computing power mean that the threats from database reconstruction attacks (DRAs) are becoming easier to realise. While the ABS has been using perturbation methods to ensure confidentiality there is a strong need to continue assessing the probability of successful DRAs and improving ABS perturbation methodology to detect and prevent DRAs. The aim of this project is to develop prototype tools that use constrained optimisation methods to quantify the amount of unit record data that can be retrieved from perturbed tables of aggregated statistics. . This research addresses the following research questions:1. How can the methods be extended to address DRA risks from multiple perturbed tables with overlapping coverages?2. How can the methods be extended to assess the change in DRA risks as more tables are produced from the same unit record data?3. Can the methods be applied to aggregates involving a mixture of categorical and continuous variable?4. Are any adjustments required to apply the methods at scale – e.g. when thousands of tables are produced from a single dataset, or when trying to reconstruct millions of records from a regression model with many coefficients?

OPTIMA CIs: Peter Taylor and Kate Smith-Miles

Associates: Joseph Chien

PhD Student: Harry Macarthur