Online learning algorithm for sparse kernel partial least squares

We present an improved online learning algorithm for sparse kernel partial least squares, this algorithm improves current methods to kernel-based regression in two aspects. First, it operates online at each time step when it acquires a new input support vector, performs an update and drop out the old data to adapted process changes. Second, it effectively reduces the dimension of feature space and accelerates the speed of training. The simulation results show the improved algorithm has good predict precision and generalize ability, and particularly useful in applications requiring on-line or real-time operation.