Dr. Luigi Pontieri

Extending Process Mining techniques with additional AI capabilities to better exploit incomplete/low-level log data: solutions, open issues and perspectives.

Recent notable advancements in the area of Process Mining (PM) have provided modern organizations with a wide range of powerful process-aware log analytics techniques. Many success stories have shown, indeed, that these techniques are a powerful means for improving a business process, especially when the available log provides good-quality information and the process is regular enough.

This keynote is primarily meant to illustrate some strategies (and concrete applications of them) for extending traditional PM techniques with further knowledge representation/learning capabilities that can help deal better with settings where the information offered by the given log data is either incomplete or too low-level.

A first strategy basically consists in exploiting (partial/incomplete) domain knowledge as a form of guidance for certain PM tasks. This strategy can turn useful, e.g., when trying to check the compliance of low-level log traces to high-level models/rules, or when facing discovery tasks against insufficient training data.

A different strategy consists in devising a sort of multi-task discovery method for automatically grasping some structured-enough (e.g., clustering-based) representation of a log’s behavior, which can help distillate better knowledge/decisions even when the process under analysis is rather complex and/or the traces in the log are at a rather low abstraction level.

The keynote will finally overview the usage of deep (reinforcement) learning solutions for predictive and prescriptive process analytics (while possibly explaining the resulting predictions/decisions), and the emerging perspective of (AI-)Augmented Analytics.

Prof. Avidgor Gal

The changing roles of humans and algorithms in (process) matching

Historically, matching problems (including process matching, schema matching, and entity resolution) were considered semiautomated tasks in which correspondences are generated by matching algorithms and subsequently validated by human expert(s). The role of humans as validators is diminishing, in part due to the amount and size of matching tasks.

Our vision for the changing role of humans in matching is divided into two main approaches, namely Humans Out and Humans In. The former questions the inherent need for humans in the matching loop, while the latter focuses on overcoming human cognitive biases via algorithmic assistance. Above all, matching requires unconventional thinking demonstrated by advanced machine learning methods to complement (and possibly take over) humans role in matching.