Learning from Zero, but Not Starting From Zero with CodaLab Worksheets

Опубликовано 22 июля 2016, 17:41

In the first half of this talk, we ask the following question: Can we learn if we start with zero examples, either labeled or unlabeled? This scenario arises in new user-facing systems (such as virtual assistants for new domains), where inputs should come from users, but no users exist until we have a working system, which depends on having training data. I discuss recent work that circumvent this circular dependence by interleaving user interaction and learning. In the second half of the talk, I will present CodaLab Worksheets (live at worksheets.codalab.org ), a platform that we have been building to help researchers manage their experiments in a reproducible way so that others can easily build off the work. CodaLab achieves this by allowing users to upload code, data, and run experiments (any shell command is permitted). CodaLab keeps the full provenance and allows the user to visualize, document the results, and create executable papers.