Adaptive Anomaly Detection for Internet of Things in Hierarchical Edge Computing: A Contextual-Bandit Approach

The advances in deep neural networks (DNN) have significantly enhanced\nreal-time detection of anomalous data in IoT applications. However, the\ncomplexity-accuracy-delay dilemma persists: complex DNN models offer higher\naccuracy, but typical IoT devices can barely afford the computation load, and\nthe remedy of offloading the load to the cloud incurs long delay. In this\npaper, we address this challenge by proposing an adaptive anomaly detection\nscheme with hierarchical edge computing (HEC). Specifically, we first construct\nmultiple anomaly detection DNN models with increasing complexity, and associate\neach of them to a corresponding HEC layer. Then, we design an adaptive model\nselection scheme that is formulated as a contextual-bandit problem and solved\nby using a reinforcement learning policy network. We also incorporate a\nparallelism policy training method to accelerate the training process by taking\nadvantage of distributed models. We build an HEC testbed using real IoT\ndevices, implement and evaluate our contextual-bandit approach with both\nunivariate and multivariate IoT datasets. In comparison with both baseline and\nstate-of-the-art schemes, our adaptive approach strikes the best accuracy-delay\ntradeoff on the univariate dataset, and achieves the best accuracy and F1-score\non the multivariate dataset with only negligibly longer delay than the best\n(but inflexible) scheme.\n