Search papers, labs, and topics across Lattice.
1
0
2
1
Shapley-based feature attribution for multi-output models *must* decompose component-wise across outputs to satisfy standard fairness axioms, revealing a fundamental constraint on joint-output explanations.
