|Ph.D Student||Wasserkrug Segev|
|Subject||A Language and Execution Model for the Inference of|
Uncertain Events in Active Systems
|Department||Department of Industrial Engineering and Management||Supervisors||Dr. Opher Etzion|
|Professor Avigdor Gal|
|Full Thesis text|
In recent years, there has been a growing need for event driven systems intended to act automatically to occurrences known as events. The earliest event driven systems were active databases. Currently, however, such active functionality is required in a variety of domains, including Business Process Management (BPM) applications, which lie at the core of enterprise's IT systems.
While some events are generated externally, other events need to be inferred by the event driven system itself based on available information sources. In many cases, such event inference is hampered by uncertainty that may be attributed to unreliable data sources and network, or the inability to determine with certainty whether an event has occurred. This issue, of event inference under uncertainty, is the issue addressed in this thesis.
In a solution to this problem of event inference under uncertainty, there are several issues which must be addressed. First, a method is required for representing the relationship between the available data, and the inferred events. Moreover, such a representation method must enable specifying the uncertainty associated with the inferred events. Second, it must be possible to calculate the uncertainty associated with each inferred event, based on some formal uncertainty representation framework. Finally, in active systems, event inference should be performed efficiently. To enable this, an efficient algorithm is required for event inference under uncertainty.
The work in this thesis has the following components: First, probability theory is chosen to quantify the uncertainty associated with the events. To represent the relations between the available data and the inferred event a probabilistic rule based formalism is used. Such a rule base approach has several major advantages, such as being intuitively appealing, and being a natural extension of deterministic event composition systems, which are also rule based. To infer the probabilities associated with the inferred events, we provide an algorithm for constructing a Bayesian Network from a set of rules and events. Although this Bayesian Network construction enables exact calculation of the associated probabilities, it is relatively computationally expensive. Therefore, we also provide a Monte Carlo sampling algorithm, which can approximate inferred event probabilities in an efficient manner. Finally, we provide results from experiments on both the sampling algorithm and Bayesian Network construction algorithm.
To summarize, our contribution in this work is the introduction of a novel, generic, and formal mechanism and framework for managing events under uncertainty conditions.