Traditional machine learning methods are always formulated by l_2 - norm distances, which are prone to outlier data and not robust against outliers. To develop a robust machine learning method, we propose a new TWSVM method measured by l_(2,p) - norm distances. The formula for this new method is difficult to solve because it needs to solve non-smooth optimization problems. In response to this problem, we provide an iterative algorithm in this article. This algorithm is proved to be strictly convergent and computationally feasible. We have done a lot of experiments on many benchmark datasets. The results show that our proposed algorithm not only has high precision, but also has better robustness.
展开▼
