Deep into Hypersphere: Robust and Unsupervised Anomaly Discovery in Dynamic Networks

Abstract

The increasing and flexible use of autonomous systems in many domains -- from intelligent transportation systems, information systems, to business transaction management -- has led to challenges in understanding the "normal" and "abnormal" behaviors of those systems. As the systems may be composed of internal states and relationships among sub-systems, it requires not only warning users to anomalous situations but also provides "transparency" about how the anomalies deviate from normalcy for more appropriate intervention. We propose a unified anomaly discovery framework "DeepSphere" that simultaneously meet the above two requirements -- identifying the anomalous cases and further exploring the cases' anomalous structure localized in spatial and temporal context. DeepSphere leverages deep autoencoders and hypersphere learning methods, having the capability of isolating anomaly pollution and reconstructing normal behaviors. DeepSphere does not rely on human annotated samples and can generalize to unseen data. Extensive experiments on both synthetic and real datasets demonstrate the consistent and robust performance of the proposed method.

Activism via attention: interpretable spatiotemporal learning to forecast protest activities

“…There have been studies that utilize the spatial, temporal or spatiotemporal dependencies in modeling or predicting the events. Several studies employed logistic regression or heuristics to forecast/detect events from social media related to anomalies [20,21], crime [22] and civil unrest [23,24]. Cadena et al [25] proposed an event forecasting model for civil unrest that uses a notion of activity cascades derived from the Twitter communication networks.…”

Deep detector classifier (DeepDC) for moving objects segmentation and classification in video surveillance

“…The first step of our proposed approach consists of segmenting foreground objects from video sequences using a deep learning‐based technique called DeepSphere [1]. Our method exploits both background subtraction and DeepSphere techniques to segment foreground objects.…”

“…In this study, we propose an idea of adapting and validating DeepSphere to the task of foreground objects segmentation in video surveillance applications. DeepSphere proposed in [1] aims at both identifying anomalous cases and exploring the cases' anomalous structure in dynamic networks localised in spatial and temporal contexts. DeepSphere exploits deep autoencoders and hypersphere learning methods to isolate pollution from anomalies and reconstruct normal behaviours.…”

“…In DeepSphere, a sample case is divided into a series of matrices corresponding to a sequence of graphs as presented in [1]. They represent the inputs for an LSTM encoder [64] and a series of internal states can be generated.…”

“…To overcome the previous problems, in this study, we take advantage of the power of an unsupervised anomaly discovery framework called DeepSphere proposed by Teng et al . [1] and adapt it to perform background subtraction and moving objects segmentation task. DeepSphere aims mainly to extract the anomalous cases in dynamic networks as well as isolating anomaly pollution and it reconstructs normal behaviour from a determined image and its corresponding background.…”