Questions Regarding Evaluation in UCL

[pgh2874]

Dear Author,

Thank you for your excellent work! I have a few questions regarding the evaluation process at UCL:

After the representation learning phase, the paper mentions measuring performance using the KNN classifier. How is it possible to predict the class without class labels using KNN? In the cited paper, Unsupervised Feature Learning via Non-Parametric Instance Discrimination, KNN voting was possible because the data had labels. How were class votes assigned in UCL, where class labels are not provided?

Thank you in advance for your time and insights!

Best regards,
Kearney