UPracticeML

The project UPracticeML aimed to expand the international master’s program “Cognitive Systems: Language, Learning and Reasoning” at the University of Potsdam (UP) in a sustainable manner by adding practice-oriented courses in the field of machine learning (ML) and new teaching formats for deep learning - including the provision of the necessary hardware. An innovative teaching concept based on research-based learning and flipped classrooms was designed to combine complex theoretical principles with practical applications such as speech recognition and text analysis. A transfer network involving partners from research, industry, and public institutions provided students with real-world problems to solve. In several consecutive practical phases, the students were supported technically and didactically by a mentoring program in order to achieve the highest possible gain in competence.

The project also enabled the establishment of a new GPU compute cluster. The goal was to provide all students participating in practice-oriented courses on deep learning with access to a graphics card on a GPU compute server so that they could work under realistic conditions. For team projects, the relevant resources were to be pooled. GPUs that were not in use at the time — for example, during semester breaks — were made available for ML research projects via a job scheduling service with minimal administrative effort. The finished cluster comprised 10 compute nodes, each with 8 Nvidia 1080Ti GPUs, as well as a central storage node and an administration server.

TODO GPU cluster image

final project report