Early Detection of Potential Risks in ICT Systems"DeAnoS: Deep Anomaly Surveillance” Deep Learning-based Anomaly Detection TechnologyNTT Network Service Systems Laboratories

As part of ongoing Network-AI and AI-OpS^*research, NTT Laboratories has developed the “’DeAnoS: Deep Anomaly Surveillance’ deep learning-based anomaly detection technology” for early detection of potential risks in ICT systems. DeAnoS efficiently analyzes the large amounts of data collected from ICT systems, and detects anomalies as deviations from a normal state. In addition, data contributing to the anomaly is inferred to be a causal factor for the anomaly.

1. ^*Application of AI to the operation of networks and related systems

Conventionally, anomaly detection in ICT systems was done by setting rules such as thresholds based on specifications and the experience of maintenance personnel, so unexpected anomalies could not be detected, sometimes leading to large-scale failures. In addition, setting rules such as thresholds for big data collected from various devices required an enormous amount of time and effort.

• Deep learning automatically learns the complex state of networks, servers, and component devices during normal operation in order to detect silent failures that would be missed by rule-based detection.
• By identifying data that is the main cause of anomalies, it reduces the time required to isolate causes shortens the period of impact on service.
• False positives are minimized even with missing or frequently changing data, and operators are not bothered by frequent alerts.

• Reducing service impact time due to failures and congestion by supporting early detection of network and servers anomalies and isolation of causes
• Implementing preventive maintenance by tracking aging related changes and contributing factors, from changes in the long-term trends of anomalies in various equipment components (e.g., HDDs).
• Implementing control that avoids anomalies detected early, by combining this technology with control technology currently being studied

NTT Network Service Systems Laboratories – Traffic, Quality and Operations Research Project