Steven Purtzel

PhD student, Databases and Information Systems, Data Systems
Humboldt-Universität zu Berlin and the Hasso-Plattner-Institute (HPI)
I am fortunate to be supervised by Prof. Dr. Matthias Weidlich

I am interested in various forms of sequential pattern matching over streams: complex event processing (CEP) and regular expressions over centralized and distributed event streams, row pattern recognition over evolving tables with SQL MATCH_RECOGNIZE, and regular path queries over streaming graphs.

In all of these settings, maintaining the state of the resulting continuous queries, in the form of partial matches, is computationally expensive and challenging, since systems face low-latency and high-throughput requirements. In response, my research investigates how to organize state information efficiently to meet these demands. For instance, when decision-making does not require all individual matches, but solely trend information, a compact stream summary suffices: a carefully selected substream with aggregated state, whose size can be tuned to the available resources. When data is produced by distributed event sources, messages need to be exchanged in an event network to detect matches. To reduce these transmissions, we have developed evaluation strategies ranging from push-pull communication to in-network processing, as well as distributed summaries that serve aggregates on demand. For parallel CEP, we have further explored query decomposition guided by the sustainable throughput of the processing units.

I very much enjoy collaborations, and most of my work has grown out of them. Recently, they have also sparked my interest in the analysis of tabular data in data lakes. Such lakes are commonly treated as static objects: what matters is mainly the current state of a table. In joint work, we instead consider a table through its temporal evolution, that is, the sequence of its versions, and study both how to manage this versioned state efficiently in temporal data lakes and what further potential they open up. Beyond that, I like to apply tools from theoretical computer science to my research problems, especially automata theory, formal languages, and complexity theory.