现有的异常检测方法在训练检测模型时绝大多数使用完全正常的样本或未标记的样本大部分是正常样本。这些方法存在的问题是：由于缺乏对异常的认识它们区分正常样本和异常样本的能力较弱。此外在研究小样本异常检测问题中发现仅使用弱监督框架训练检测模型对训练样本中的异常噪声或新的数据非常敏感。因此为了提高此模型的抗噪声能力提出了一种基于相关熵Correntropy Criterion的小样本异常检测方法。利用信息理

Most existing anomaly detection methods for training detection models use fully normal samples or unlabeled samples (mostly normal samples). The problem with these methods is that their ability to distinguish normal samples from abnormal samples is weak due to the lack of knowledge about anomalies. In addition, it has been found in the study of small sample anomaly detection problems that training detection models only using weakly supervised frameworks is very sensitive to anomalies (noise or new data) in the training samples. Therefore, to improve the noise resistance of this model, a small sample anomaly detection method based on Correntropy Criterion is proposed. By introducing the Correntropy Criterion from information theory learning into the reconstruction of anomaly detection, a new framework is constructed. The width parameter of the Correntropy function can be adjusted to effectively suppress the adverse effects of noise. Meanwhile, the optimization problem of the proposed method is solved using semi-quadratic optimization technology, requiring only a few iterations to obtain a local optimal solution. Experimental results show that the proposed method has better noise resistance and generalization performance