EVIDENTIAL REASONING FRAMEWORK

Evidential Reasoning Framework Capability Brief
Evidential Reasoning Framework Capability Brief

Many applications rely on accurate and frequent readings acquired from physical and logical sensors. However for many deployments the sensor data cannot be deemed accurate and it can be characterised by uncertainty or incompleteness.

This ‘noise’ is often introduced into analytic sensing where imperfect analytic algorithms cannot deliver dependable accuracy in all cases (consider for example video analytics). Nevertheless when the level of uncertainty can be quantified, significant benefits result from reasoning with this uncertainty over extended time periods.

Researchers at CSIT have extended traditional Artificial Intelligence reasoning techniques and incorporated state-of-the-art uncertainty and conflict handling to create a powerful Evidential Reasoning Framework (ERF) for use across a broad range of applications. Application opportunities are numerous but within CSIT the framework has been applied to traditionally difficult problems such as autonomous surveillance.

A multi-agent architecture allows reasoning to be devolved to semiautonomous agent groups, distribute computation and deliver scalability and fault tolerance beyond that available via centralised approaches. A rich user interface allows users to create and config- ure agents and to define agent behaviour through a rule builder utility. A toolkit allows developers to integrate unsupported sensors.

The ability to reason uncertain, conflicting and incomplete event data allows crucial composite events to be inferred and detected where they would otherwise have been missed or interpreted incorrectly using traditional methods. This is particularly important for surveillance applications where complex and busy environments present frequent periods of uncertainty or for applications such as Network Traffic Analysis where real-time performance can only be obtained by intermittent sensing.

The full capability brief is available to download from our website here - CB3-ERF (Evidential Reasoning Framework) (pdf, 1.3MB)