Last week, I joined the 5-days Schloss Dagstuhl Seminar on Data Series Management organized by prominent researchers in this field, including my Ph.D. advisor Themis Palpanas. I didn’t know what to expect but I had no doubts about the quality of the participants.
With more than five research publications on modeling and querying data series, and two years in the industry working on spatiotemporal data analytics, I was intrigued to see where this was heading to!
First, some context: Schloss Dagstuhl is a research center for computer science located in a very remote and relaxed area in the countryside of western Germany providing accommodation, catering, conference and meeting rooms, a comprehensive library with more than 50,000 volumes, and many amenities whose ultimate goal is to maximize interactions between the attendees. In addition, the DBLP bibliographic service is managed by the permanent staff of Schloss Dagstuhl: if you’ve been involved in academic research on computer science, you’ve been to Schloss Dagstuhl — either physically or digitally. 😁
The format is simple: throw a bunch of really smart people in a remote place, get them to talk about their work and challenges, and consolidate the discussion into a final report. The organization and schedule have been tuned for decades:
Participants arrive on Sunday and leave on Friday afternoon
Monday starts off with a few words from the organizers: intro, motivation, logistics, and agenda overview. There’s a round of 2-minute introductions of the participants, followed by a second round of 5-minute lightning talks about relevant research and challenges. The day ends with an open discussion on the main directions of interest
From Tuesday to Thursday morning, the work is organized in groups covering 3-5 parallel breakout sessions lasting three hours each. At the end of each session, there’s a plenary 5-minutes presentation with Q&A for each group
Wednesday afternoon it’s time for the social event: we visited Trier, the last bastion of the Roman Empire’s presence north of the Alps
Thursday afternoon and Friday morning is devoted to writing collaboratively the final report, later used as the basis for one or more other research publications
The loose organization encourages the creative process but can also become just a big mess with very low productivity. The following tricks worked pretty well to foster creativity without destroying productivity:
Setting clear goals from day one helps to guide the discussion toward an expected convergence point
Attendees are encouraged to change groups several times: this helps to spread knowledge and ideas. E.g., an expert in visualization can join the “visualization” group to share some novel methods and the “query evaluation” group to discuss some visualization requirements
Taking meeting minutes of each breakout session is very important to reason and consolidate the progress, providing the content for the plenary group presentations and for the final report
The draft of the final report can be created collaboratively with online LaTeX editors like Overleaf. It was amazing to see a draft growing so fast with more than twenty people editing it in parallel
Seating at lunch and dinner is randomized to increase your chances to meet new people using small tables of five seats each. Worth mentioning, dinner is served at 18:00, giving plenty of time afterward for a few drinks while playing pool or for a walk to the old castle.
The lightning talks covered a wide range of novel problems and applications: surgical robotics, aerospace, seismology, wind farm data management, augmented healthcare, location intelligence, federated interactive data visualization, anomaly detection, classification with deep learning, imputation of missing values, managing and querying massive datasets, progressive similarity search, representation learning, progressive PCA, higher classification accuracy with refusals, visualisation of large datasets, socio-temporal data mining.
I really enjoyed meeting old friends and making new ones. We came out with a strong analysis of the current challenges in data series management that will shape the future of this increasingly important research area motivated by clear industry requirements.