Episodic Memory Representation in a Knowledge Base , with Application to Event Monitoring and Event Detection

The thesis explores the use of a knowledge-based AI system to assist people in executing procedures and detecting events in an observed world. We extend the Scone knowledge-base system with open-ended facilities for representing time and events. Then we use this episodic knowledge representation as the foundation for our event monitoring and event detection algorithms. This approach lets us represent and reason about three fundamental aspects of the observed events: 1. their ontological character and what entities take part in these events (e.g. buying is a kind of transaction that involves an agent, a seller, money and goods) 2. how events change the world over time (e.g. after a buy event the agent has the goods rather than the money) 3. how events may be composed of other subevents (i.e. a buy event may be composed of giving money and receiving goods) We illustrate knowledge-based solutions to the event monitoring problem in the conference organization domain and to the event detection problem in the national security domain.