Anomaly detection aims to identify the true anomalies from a given set of data
instances. Unsupervised anomaly detection algorithms are applied to an unlabeled
dataset by producing a ranked list based on anomaly scores. Unfortunately, due to
the inherent limitations, many of the top-ranked instances by unsupervised algorithms
are not anomalies or not interesting from an application perspective, which leads to
high false-positive rates. Active anomaly discovery (AAD) is proposed to overcome
this deficiency, which sequentially selects instances to get the labeling information and
incorporate it into the anomaly detection algorithm to improve the detection accuracy
iteratively. However, labeling is often costly. Therefore, the way to balance detection
accuracy and labeling cost is essential. Along this line, this paper proposes a novel AAD
method to achieve the goal. Our approach is based on the state-of-the-art unsupervised
anomaly detection algorithm, namely, Isolation Forest, to extract features. Thereafter,
the sparsity of the extracted features is utilized to improve its anomaly detection performance. To enforce the sparsity of the features and subsequent improvement of the
detection analysis, a new algorithm based on online gradient descent, namely, Sparse
Approximated Linear Anomaly Discovery (SALAD), is proposed with its theoretical
Regret analysis. Extensive experiments on both open-source and additive manufacturing datasets demonstrate that the proposed algorithm significantly outperforms the
state-of-the-art algorithms for anomaly detection.
3D Printing in Heterogeneous Catalysis—The State of the Art
