Some questions about the CIFAR-100 experiment

[FutureXiang]

Dear authors,

Thank you for sharing this amazing and inspiring work!
I am really interested in the CIFAR-100 experiment presented in your paper, and I have a couple of questions about it.

Specifically, could you please share some information about the feature extractors and clusters used for CIFAR-100?

• For example, what's the NMI value of the 400 clusters used in Figure 3?
• Do you use supervised or self-supervised pre-trained network to extract features? Could you please share the specific algorithm (e.g., MoCo or SimCLR ?) and some metrics about these extractors, such as top-1 accuracy or linear probing accuracy?

Thanks again for your great work, really looking forward to your response!