Department of
Computational
Perception
|
Special Topics: Reproducible Computational Research
General Information
Scientific progress requires independent researchers to reproduce published results to verify them. Such verification should be easy in computational sciences like machine learning, where scientists could freely share data and program code. However, current research practice often does not allow reproducing results easily.
This course shall give students an understanding of the importance and difficulties of reproducing computational research, as well as introduce students to tools that facilitate it.
Course Content
The course will consist of lectures as well as a practical part. The lectures will introduce tools and best practices for reproducible computational research that the students can use in the practical part, e.g. version control, keeping track of experiments, making sure the code is correct, etc. In the practical part, students will choose a scientific paper, try to reproduce the results, and create a "reproducibility package" that enables other students to reproduce their results easily. At the end of the course, students will present these results, as well as describe problems they faced (and how they solved them) in this process.