Open Ako-r opened 1 year ago
Hello @Ako-r , in NCD we are given an unlabeled dataset as well as an additional labeled dataset. The labeled dataset has different but related classes to the unlabeled one. This is the main difference with clustering. So the goal in NCD is to use the labeled set to guide the clustering process to find more relevant classes than you could with unsupervised clustering.
Hope this clears the difference for you.
Hi! Thank you for organizing the relevant work. I am confused about the differences between novel class discovery and unsupervised learning. Would it be convenient for you to answer for me? Thanks!