PROJECT TITLE :

Kernel Recursive Least-Squares Tracker for Time-Varying Regression

ABSTRACT:

During this paper, we have a tendency to introduce a kernel recursive least-squares (KRLS) algorithm that is in a position to track nonlinear, time-varying relationships in information. To this purpose, we first derive the quality KRLS equations from a Bayesian perspective (as well as a wise approach to pruning) and then exploit this framework to incorporate forgetting in an exceedingly consistent way, thus enabling the algorithm to perform tracking in nonstationary eventualities. The resulting methodology is the first kernel adaptive filtering algorithm that features a forgetting factor in an exceedingly principled and numerically stable manner. Over and above its tracking ability, it's a variety of appealing properties. It is online, requires a fixed quantity of memory and computation per time step, incorporates regularization in an exceedingly natural manner and provides confidence intervals together with each prediction. We embrace experimental results that support the idea with illustrate the efficiency of the proposed algorithm.