Lifting Weak Supervision To Structured Prediction

This paper finds that weak supervision can be used beyond classification applications, including rankings, graphs, and manifolds, and can provide generalization guarantees nearly identical to models trained on clean data.