Assume there are \(C\) relevance grades from \(0\) to \(C-1\) in the training data, and we employ the following two noise injection methods respectively to change its relevance labels. Uniform noise profile: randomly change some of the relevance labels to other labels uniformly, i.e. ...