Things to try on CAL

[guneytombak]

Change number of neighbors (50 or 3)
Change distance metric (l2 or symmetric kl)
Use the index instead of probability (~KNN)
Inverse the counting (might work or might worsen)

[AronDiSc]

• resolve initialization issue @guneytombak
• different population size (e.g. 1%, 2%, 3%) @guneytombak
• l2 distance @guneytombak
• Pretrained VAE for MNIST, fashion MNIST (CIFAR100) @AronDiSc
• Dataloader MNIST, fashion MNIST @remo48
• Low dimensional representation -> plot / visualization https://scikit-learn.org/stable/modules/generated/sklearn.manifold.TSNE.html @Sebi1128
  => Until Friday 31/12/21