关键词: 主动学习, 高斯过程回归, 协方差矩阵, 相似性

Abstract:

Since the data collected from industrial processes often contain a large number of unlabeled samples, while the number of labeled samples is small and the cost of manual labeling is high, an active learning method based on covariance matrix is proposed. This method uses labeled samples to establish a Gaussian process regression model, and constructs the covariance matrix between the unlabeled samples, using the value of the determinant of the covariance matrix as an evaluation indicator. While selecting informative unlabeled samples, the similarity between samples is measured to avoid redundant addition of samples, which finally improves model prediction accuracy at the same labeling cost. The application simulation of the algorithm based on industrial process data verifies the feasibility and effectiveness of the proposed method.

Key words: active learning, Gaussian process regression, covariance matrix, similarity words