Supplementary Material: The Role of Global Labels in Few-Shot Classification and How to Infer Them

The supplementary material is organized as follows: Appendix A contains the proofs accompanying our theoretical analysis. Appendix B presents additional experiment results. For a dataset D, let π (D) be the set of class labels from D . In (A.4), the inequality is formed because the denominator MeLa is compatible with different meta-learning algorithms. Table 7 suggests that MeLa obtains test performance comparable to RFS, which is the oracle setting.