Residual k-Nearest Neighbors Label Distribution Learning

Jing Wang, Fu Feng, Jianhui Lv, Xin Geng

Abstract: Highlights•Residual label distribution learning. We propose residual label distribution and design residual label distribution learning.•Neighborhood structure of label distribution. We propose to exploit the neighborhood structure of label distribution.•The RkNN-LDL method. We put forward a novel LDL method called residual kNN label distribution learning (RkNN-LDL).•Generalization and experiments. We prove an O(m/n) generalization bound for RkNN-LDL and validate its performance in experiments.