Conducting research in the areas of data engineering, data management, and machine learning systems requires the ability to deal with scientific literature in these areas as well as to design, implement, and evaluate prototypes. To facilitate these skills, the DAMS Lab group (FG Big Data Engineering) at TU Berlin offers a seminar and a project on Large-scale Data Engineering as a combined module (12 ECTS), which can be taken by bachelor and master students. Taking both seminar and project is the ideal preparation for a bachelor/master thesis with our group. Alternatively, only bachelor students may take the seminar as a separate module (3 ECTS) and the project as a separate module (9 ECTS).
Modules and assigned degree programsIn the beginning of the semester, students will hear introductory lectures on reading scientific papers, finding related work, writing high-quality scientific papers, and giving a high-quality scientific presentation. Each student selects a topic, reads and understands the given paper, searches for related work, and writes a short summary of the assigned paper. In the end of the semester, each student gives a slide presentation in front of the group.
This semester's umbrella topic: Efficiently Combining DB and ML Workloads
Database query processing and ML training and scoring are normally executed in dedicated systems. However, there is a trend towards integrated data analysis pipelines involving both query processing and ML. Unfortunately, the orchestration of existing DB and ML systems is inefficient due to expensive data transfer and missed global optimization potential. This semester, we deal with recent research papers addressing these challenges through: (a) improving the data transfer between DB and ML systems, (b) running one kind of workload on existing software/hardware designed for the other kind of workload, (c) creating entirely new systems supporting both query processing and ML at the same time. These solutions affect all levels of the system stack, from query languages over optimization and compilation techniques as well as local/distributed runtime techniques to the use of multi-core CPUs and hardware accelerators.
List of topics: Separate list of topics
Submission & deadlinesIn the beginning of the semester, students/teams pick a programming project from a provided list, devise an initial design and then implement a prototype including documentation, tests, and relevant experiments. The project ends with a presentation of the obtained results in front of the group.
The topics of the project are independent of the seminar. We will offer tasks in a wide range of components of data management and machine learning systems. Each individual project will be conducted in the context of one of the two systems developed by our group (and other collaborators) as part of our research:
Thereby, students get the chance to make meaningful contributions to free open-source projects. The projects can be done either individually or in teams of up to three students (with the expected amount of work proportional to the team size).
List of topics: Separate list of topics (updated Oct 20)
Submission & deadlines