Comparison theorems on large-margin learning

This paper studies the binary classification problem associated with a family of Lipschitz convex loss functions called large-margin unified machines (LUMs), which offers natural bridge between distribution-based likelihood approaches and margin-based approaches. LUMs can overcome so-called data piling issue support vector machine in high-dimension low-sample size setting, while their theoretical analysis from perspective learning theory is still lacking. In this paper, we establish some new comparison theorems for all LUM play key role error algorithms. Based on obtained theorems, further derive rates regularized schemes varying Gaussian kernels, maybe independent interest.