ECAI 2014 Tutorial: Formal Methods for Event Processing

 

Tutorial Description

Organisations require techniques for automated transformation of the Big Data they collect into operational knowledge. This requirement may be addressed by employing event processing systems that detect activities/events of special significance within an organisation, given streams of low-level information that are very difficult to be utilised by humans.

We will review a Chronicle Recognition System (CRS), the Event Calculus (EC), ProbLog and Markov Logic Networks (MLN). CRS is a purely temporal reasoning system that allows for efficient event processing. EC allows for the representation of temporal and atemporal constraints. Consequently, EC may be used in applications requiring spatial reasoning, for example. ProbLog and MLN, unlike EC and CRS, allow for uncertainty representation and are thus suitable for event processing in noisy environments.

The manual development of event structures is a tedious, time-consuming and error-prone process. Moreover, it is often necessary to update such structures during the event recognition process, due to new information about the application under consideration. For this reason, we will review machine learning techniques automating the construction and refinement of event definitions.

To illustrate the reviewed approaches we will use real-world case studies from the FP7 SPEEDD project: event processing for city transport and traffic management, and credit card fraud management.

Moreover, we will give a demo of each approach, upon request, using one of the event recognition datasets that we have at our disposal.

Syllabus

The slides of the tutorial are available here. A paper on the tutorial is available here.

Intended Audience

The intended audience of the tutorial consists of academics, students and practitioners investigating the open issues of event processing, and/or willing to apply event processing techniques for extracting knowledge from structured and unstructured datasets. Familiarity with AI techniques is desirable

Presenters

Alexander Artikis is a research associate in NCSR Demokritos (Athens, Greece). He holds a PhD from Imperial College London on norm-governed multi-agent systems, while his research interests lie in the areas of AI and distributed systems. Alexander has been working on several international projects on event processing; currently, he is the technical director of the FP7 SPEEDD project that develops a Big Data system for proactive event-driven decision-making. Alexander has also developed a highly scalable, logic-based, open-source event processing system.

Georgios Paliouras is a senior researcher in NCSR Demokritos (Athens, Greece). He holds a PhD from Manchester University on machine learning for event recognition. He has performed basic and applied research in machine learning for the last 17 years. He is involved in many European and national research projects and has the role of scientific coordinator in some of them, including the FP7 SPEEDD project. He has given a number of invited talks and tutorials at various institutions and conferences, such as ECML/PKDD, DEBS and IJCAI.

Alexander Artikis