Continuous GNN-based Anomaly Detection on Edge using Efficient Adaptive Knowledge Graph Learning

TL;DR

This work tackles the challenge of robust Video Anomaly Detection (VAD) in edge and resource-constrained environments by enabling continuous, on-device adaptation of a MissionGNN-generated Knowledge Graph (KG). It introduces a three-phase KG modification pipeline (pruning, altering, creating nodes) and a lightweight GNN-based decision model that, together with a transformer-based temporal component, operates entirely on edge devices without cloud updates. A key contribution is the continuous KG adaptive learning mechanism, which updates only KG token embeddings in response to evolving anomaly patterns, guided by a dynamic selection of top anomalies and a defined update rule. An interpretable KG retrieval component translates updated embeddings into human-readable concepts, aiding interpretability, while experimental results on UCF-Crime demonstrate sustained performance with reduced cloud dependency and improved practicality for dynamic, real-world deployment.

Abstract

The increasing demand for robust security solutions across various industries has made Video Anomaly Detection (VAD) a critical task in applications such as intelligent surveillance, evidence investigation, and violence detection. Traditional approaches to VAD often rely on finetuning large pre-trained models, which can be computationally expensive and impractical for real-time or resource-constrained environments. To address this, MissionGNN introduced a more efficient method by training a graph neural network (GNN) using a fixed knowledge graph (KG) derived from large language models (LLMs) like GPT-4. While this approach demonstrated significant efficiency in computational power and memory, it faces limitations in dynamic environments where frequent updates to the KG are necessary due to evolving behavior trends and shifting data patterns. These updates typically require cloud-based computation, posing challenges for edge computing applications. In this paper, we propose a novel framework that facilitates continuous KG adaptation directly on edge devices, overcoming the limitations of cloud dependency. Our method dynamically modifies the KG through a three-phase process: pruning, alternating, and creating nodes, enabling real-time adaptation to changing data trends. This continuous learning approach enhances the robustness of anomaly detection models, making them more suitable for deployment in dynamic and resource-constrained environments.

Figures (6)

• Figure 1: Demonstration of how our proposed framework dynamically adapts to anomaly trend shift in the real world.
• Figure 2: The overall framework for our proposed KG adaptive learning framework utilizing the novel concept of hierarchical graph neural network.
• Figure 3: Detailed description of our mission-specific knowledge graph generation framework.
• Figure 4: Pipeline of our proposed knowledge graph modification mechanism.
• Figure 5: Test AUC scores measured across shifts in the anomaly target, comparing different types of anomalies.