Personalized Federated Learning with Semisupervised Distillation

Figure 6

In the case of homogeneous and heterogeneous models, the influence of different public dataset sizes on the test performance of (a, b) pFedSD, and (c, d) all methods on MNIST with  = 10, frac = 0.8, and  = 100. (a) Homo-models (pFedSD). (b) Hetero-models (pFedSD). (c) Homo-models (all). (d) Hetero-models (all).

(a)

(b)

(c)

(d)