Some of the interesting work we do over at the Videorooter project is automated testing of algorithms. In short, we're ranking algorithms for visual matching of images and videos as per how well they perform. High accuracy and low number of false positives give a high mark. What's new is that we've connected this to the Github repositories, so when someone forks our code and changes it (potentially improving the algorithm), the automated tests are run.. well, automatically, and the performance of the new algorithm automatically included in our statistics.
Read more in my blogpost about testing algorithms over at the Videorooter project to learn more.