Humboldt-Universität zu Berlin - Mathematisch-Naturwissenschaftliche Fakultät - Software Engineering

Software Engineering für Bilderkennungs ML Pipeline


 

Wann und Wo

Semesterprojekt: Dienstag, 13-17 Uhr, RUD 26, 1'305

Wer

Dozent: Eik Reichmann / Marc Carwehl

Beschreibung und Aufbau der Lehrveranstaltung

Due to the recent improvements in AI algorithms and the increase in compute power, the demand and adoption of AI-based functionalities has dramatically increased. However, developing AI-based systems is not only about the best possible algorithm but more about a reliable training and testing pipeline that process data, trains models, and evaluates results in a reproducible way.

This semester project focuses on the practical aspects of engineering AI systems. Students will implement a complete pipeline, including data gathering and preprocessing as well as model training and evaluation. The task is image recognition. The students are encouraged to apply software engineering practices such as modular design, version control, reproducibility, parallel training and testing to their ML pipeline. To introduce some additional motivation students will compete against each other in the development of the best model.
In the end they will have to write a short report about their approach.

 

Organisatorisches

Die Lehrveranstaltung findet auf Englisch statt.