Machine Learning Based System Identification with Binary Output Data Using Kernel Methods

Abstract

Within the realm of machine learning, kernel methods stand out as a prominent class of algorithms with widespread applications, including but not limited to classification, regression, and identification tasks. Our paper addresses the challenging problem of identifying the finite impulse response (FIR) of single-input single-output nonlinear systems under the influence of perturbations and binary-valued measurements. To overcome this challenge, we exploit two algorithms that leverage the framework of reproducing kernel Hilbert spaces (RKHS) to accurately identify the impulse response of the Proakis C channel. Additionally, we introduce the application of these kernel methods for estimating binary output data of nonlinear systems. We showcase the effectiveness of kernel adaptive filters in identifying nonlinear systems with binary output measurements, as demonstrated through the experimental results presented in this study.