New Insights Into Learning With Correntropy-Based Regression

Stemming from information-theoretic learning, the correntropy criterion and its applications to machine learning tasks have been extensively studied explored. Its application regression problems leads robustness-enhanced paradigm: correntropy-based regression. Having drawn a great variety of successful real-world applications, theoretical properties also investigated recently in series studies statistical viewpoint. The resulting big picture is that regresses toward conditional mode function or mean robustly under certain conditions. Continuing this trend going further, study, we report some new insights into problem. First, show additive noise model, such paradigm can be deduced minimum distance estimation, implying estimator essentially thus possesses robustness properties. Second, fact provides unified approach it approaches mean, mode, median functions Third, present results when used learn by developing error bounds exponential convergence rates ([Formula: see text])-moment assumptions. saturation effect on established rates, which was observed assumptions, still occurs, indicating inherent bias estimator. These novel deepen our understanding regression, help cement theoretic framework, enable us investigate schemes induced general bounded nonconvex loss functions.