The paper mentions unsupervised learning is overshadowed by the success of supervised learning. I think this is because it is more hard to test the accuracy of unsupervised learning. Since unsupervised learning don't have response, if we use complex algorithms from deep learning, we might get some latent results that cannot be directly obtained from the row data. I want to know how can we overcome this kind of problem that it might be more hard to tell the accuracy of unsupervised learning?
The paper mentions unsupervised learning is overshadowed by the success of supervised learning. I think this is because it is more hard to test the accuracy of unsupervised learning. Since unsupervised learning don't have response, if we use complex algorithms from deep learning, we might get some latent results that cannot be directly obtained from the row data. I want to know how can we overcome this kind of problem that it might be more hard to tell the accuracy of unsupervised learning?