Disentangling Labeling and Sampling Bias for Learning in Large-output Spaces

  Ankit Singh Rawat 1 Aditya Krishna Menon 1 Wittawat Jitkrittum 1 Sadeep Jayasumana 1 Felix X. Yu 1
                  Sashank J. Reddi 1 Sanjiv Kumar 1

             Abstract                closely a scheme approximates the unsampled loss on the
                                full label set. This bias is well understood for sampled
   Negative sampling schemes enable efficient train-
                                s ...