Fast Training and Efficient Linear Learning Machine

Time complexity is a challenge for learning machines. In this paper, a fast training and efficient linear learning machine is presented. Starting from a simple linear classifier, a new one is proposed based on an improvement on the first one. The machine obtained is characterized by a weight vector that can be processed immediately without any complex calculus or optimization step, which allows for considerable training time savings. A geometric interpretation of the proposed method is given. Experiments show that this classifier is competitive to other state of the art linear learning methods such as Support Vector Machines and Kernel Fisher Discriminant.