Reproducibility

Better Documented Experiments.

Being able to reproduce and verify experimental results is a principal prerequisite of scientific research. Still, it is often hard to reproduce learning experiments, either because datasets and algorithm implementations are not available or change over time, or because paper publications don't allow to fully describe all details of the experimental setup.

Uploading all the details of machine learning experiments to an experiment database provides a standardized online log, allowing researchers to easily make the details and resuls of their experiments available to their peers. As such, it becomes much easier to reproduce and verify previous experiments.