Creating Training Sets via Weak Indirect Supervision

This paper extends the scope of usable sources in WS, by formulating Weak Indirect Supervision (WIS), a new research problem for automatically synthesizing training labels based on indirect supervision sources that have different output label spaces.