“…Wang proposes to use an upper bound L1 parametrization,
11 setting an upper limit of loss
, where the loss does not exceed
regardless of the misclassify‐cation level of the sample, to limit the value of the loss function to
ensure better robustness. Xie introduces the Lp‐norm regularization term,
12 to effectively exploit the geometric information embedded in the data, avoiding overfitting while improving the generalization ability of the algorithm. Liang modeled each uncertain sample as a random vector with a Gaussian distribution of the proposed model
13 .…”