Trusted AI Frameworks for Change and Anomaly Detection in Obsverved ISR Patterns

P2-11


Project Leader: Matthew Roughan (UoA)

Participants: SmartSat CRC Ltd, University of Adelaide, BAE Systems

This project seeks to automate the identification of higher order patterns in ISR
(Intelligence, Surveillance and Reconnaissance) sensed detections along with
establishing normalcy. The intention is for significant changes from normalcy –
anomalies – to be reported to operators as alerts requiring human assessment,
decision, and action. In addition, the rationale of the alerts will also be computed
and presented in a transparent way to instil user confidence in the results.

Two novel aspects for this project are: (1) the use of multiple strategies for
pattern detection, including deep learning and advanced statistical modelling
(e.g., Bayesian Computation); and (2) the incorporation of a Pattern Question
Answering (PQA) capability to enable intuitive interaction and interrogation of the
reported patterns for their rationale. PQA will build on and generalise existing
capabilities in Visual Question Answering (VQA) in the fields of Artificial
Intelligence and Machine Learning.

Specific application domains will be considered to support the development and
demonstration of capability, including domains such as maritime traffic, space
situational patterns, and land use patterns.