Efficient analysis method for unreliable labeled data

点击翻译按钮获取中文摘要

In the analysis of real data such as the biological signals, the given labels are often unreliable. Because, objects to be measured inherently contain some physical and biological uncertainty, and some labels might be incorrectly assigned by human intuition. Whereas, reliable labels would be available for a small portion of the samples. In such case, a semi-supervised learning method is effectively applied to analyze the data, estimating the label values of samples. In addition, it is favorable that the estimated label values provide us the degree of confidence of each sample. In this research, we proposed a novel method of semi-supervised learning, incorporating logistic functions into label propagation in order to accurately estimate the label values as the posterior probabilities. We call this method logistic label propagation(LLP). In addition, we proposed a novel optimization method for LR by directly using the non-linear conjugate gradient method in order to apply to LLP and to reduce the computational cost. Our proposed methods achieve the better estimation of degree of confidence and the faster computation times compared with the ordinary methods.