Appendices A Details of Online Few-Shot Learning Experiments

A.2 T ask Details: Figure 6 shows a sample online 10-shot 5-way task. As mentioned in Section 3.1, at time step i, we We follow the same protocol as in Santoro et al. For both OPN and CPM, we use the same features extractor as in Ren et al. Note: In the APL paper they use a ResNet as the feature extractor. In initial experiments, we observe that for tasks with 50 time steps these models did not train well.