Reproducing experiments is central in research methodology, also AI research. However, recently the research community in general has uncovered that a large amount of the published research - even research published in the most prestigious journals - is hard or impossible to reproduce, sometimes even for the researchers conducting the research to begin with.
Several tasks are open:
1) Design a natural language processing system that can analyze to which degree a research paper can be reproduced.
2) Investigate issues related to the reproducibility of different machine learning techinques.
3) Build a reproducibilty framework for the Norwegian Open AI Lab.
4) Investigate the different levels of reproducibility and what it means to reproduce results.
The task is to continue ongoing research with aims at publishing at high impact conferences and journals. We have published at AAAI, eScience and AI Magazine. Previous work has been mentioned in Science and Morgenbladet.