Adversarial Multiclass Learning under Weak Supervision
                with Performance Guarantees

         Alessio Mazzetto * 1 Cyrus Cousins * 1 Dylan Sam 1 Stephen H. Bach 1 Eli Upfal 1

             Abstract                 conflict with one another. We assume access to only a
   We develop a rigorous approach for using a set        small amount of ground-truth labeled data. Much prior
   of arbitrarily correlated weak supervision sources      work on aggregating noisy labels (Dawid & Skene ...