OrEdge: Lightweight Multi-Modal Anomaly Detection via Orthogonal-Domain Learning
A team of researchers has unveiled a novel system named OrEdge, aimed at real-time anomaly detection in multi-modal distributed software systems. This innovative approach differentiates itself from traditional methods by utilizing orthogonal-domain temporal representations, significantly lowering computational requirements and simplifying model complexity. OrEdge effectively analyzes a range of monitoring data, such as logs and metrics, to pinpoint irregular software activities and understand their temporal associations. Central to OrEdge is the OrEdgeCore module, which efficiently identifies stable temporal patterns while minimizing extraneous signals. Testing on three microservice datasets has demonstrated its notable accuracy and efficacy in larger software environments.
Key facts
- OrEdge is a lightweight framework for real-time anomaly detection in multi-modal distributed software systems.
- It uses orthogonal-domain temporal representations instead of attention- and graph-based architectures.
- It jointly analyzes logs, metrics, and traces to identify abnormal software behavior.
- OrEdgeCore is a lightweight orthogonal-domain reconstruction module that captures recurring temporal patterns.
- Evaluated on three real-world microservice datasets: MSDS, SN, and TT.
- Achieves competitive detection performance with lower computational complexity and model size.
- Reduces redundancy across observability signals.
- The framework is designed for efficient deployment in distributed systems.
Entities
—