Combinatorial Inference against Label Noise

In our framework, we automatically generate multiple coarse-grained meta-class sets, each of which constructs a heterogeneous partition of the original class set.