[eychen]

My concern is that the probability of "SVM outperforms PLA" gets higher and higher as the training set grows. I find this a bit interesting because the VC dimension of SVM seems to approach that of PLA as the training set grows.