The L2-regularized hinge loss kernel SVM could be the most important and most studied machine learning algorithm. Unfortunately, its computational training time complexity is generally unsuitable for big data. Empirical runtimes can however often be reduced using shrinking heuristics on the training sample set, which exploit the fact that non-supportâ€¦ (More)

Training of one-vs.-rest SVMs can be parallelized over the number of classes in a straight forward way. Given enough computational resources, one-vs.-rest SVMs can thus be trained on data involving a large number of classes. The same cannot be stated, however, for the so-called all-in-one SVMs, which require solving a quadratic program of size quadraticallyâ€¦ (More)