A Proof of Theorem 2 Lemma 1 (McDiarmid)

Lemma 3 are proposed in [20] to prove Theorem 3. Lemma 2 As shown in Eq. (6), the generalization bound of meta-algorithms with LOO training relies on both Stability of Inner-T ask Algorithm. Prototypical networks find the prototype (mean vector) of each class first and then classifying the query into the nearest prototype's class in the embedding space. Based on the definition, the following gives the derivation of the stability parameter β . For the expectation w.r.t. the support set Therefore, we obtain the upper bound of O (1/m) for the hypothesis stability β . Based on the above results, we obtain Theorem 5. Theorem 5