The positions are within dr. Sicco Verwer's recently funded VIDI project "Learning state machines from infrequent software traces". The project goal is research and develop novel model learning algorithms and develop tools for learning from software log data. The aim is to provide software analysts with insightful models. We are able to learn insightful state machine models for the frequent happy flow of software. Although these can be useful to understand software, the real interesting behavior occurs in the infrequent unhappy flows, typically caused by errors. The project aims at learning state machines from such unhappy flows. Using techniques from outlier explanation we aim to uncover how unhappy flows differ from happy ones (position 1). Additional information available in software log databases such as co-occurring flows will be used as a form of active learning to obtain more information about unhappy flows (position 2). We will work close together with developers and analysts from our industrial partners Adyen and Bitdefender to develop a tool that helps them to understand software/network logs and errors.
Further details:
2 PhD Students in Model Learning at Delft University of Technology