Anomaly detection is a significant task of data mining, and also a hot research topic in various fields of artificial intelligence in recent years. It has a wide range of applications, such as extreme climate event detection, mechanical fault detection, terrorist detection, fraud detection, malicious URL detection etc.
There are many kinds of classic shallow anomaly detection methods proposed to solve the problem of anomaly detection in various scenarios. However, the explosive growth of databases in both size and dimensionality is challenging for anomaly detection methods in two important aspects: the requirement of low computational cost and the susceptibility to high-dimensionality issues. Efficient methods are in high demand for time-critical applications ranging from network intrusion detection to credit card fraud detection.
Recently, deep learning has shown Phenomenal success in tackling these complexities in a wide range of applications, but popular deep learning techniques are inapplicable to anomaly detection due to some unique characteristics of anomalies, e.g., rarity, heterogeneity, unbounded nature, and prohibitively high cost of collecting large-scale anomaly data. A large number of studies, therefore, have been dedicated to deep methods specifically designed for anomaly detection. These studies demonstrate great success in addressing some major challenges to which shallow anomaly detection methods fail in different application contexts.
This tutorial aims to present a comprehensive review of both shallow and deep-learning-based anomaly detection with explanation. We first introduce the key intuitions, objective functions, underlying assumptions, and advantages and disadvantages of state-of-the-art anomaly detection methods. We also introduce several principled approaches used to provide anomaly explanations for deep detection models. Furthermore, we will discuss the connections between classic shallow and novel deep methods and provide a practical guide on how to select an outlier detector in different applications.
Contributors
Dr Ye Zhu, Deakin University, Australia
Dr Guansong Pang, Singapore Management University, Singapore
Mr Xin Han, Macau University, China
Mr Yang Cao, Deakin University, Australia
A/Prof. Mark Carman, Politecnico di Milano, Italy
Prof. Gang Li, Deakin University, Australia
Dr Yue Zhao, Carnegie Mellon University, USA
Mr Zong-you Liu, Nanjing University, China
Dr Xuyun Zhang, Macquarie University, Australia
Dr Sutharshan Rajasegarar, Deakin University, Australia
Dr Chen Li, Nagoya University, Japan
Prof. Ting Kai Ming, Nanjing University, China