Abstract Attack Intention Inference Using Low-Rank Gated Arithmetic Interactive Attention

Internal attack originated from devices in internet of things network is an essential security threat. We defined the abstract attack intention in internet of things network, and proposed an extended, interactive attention based deep architecture, that can be trained on the tabular based internet of things network traffic data to precisely identify 4 kinds of abstract attack intentions including disable target devices, gathering intelligence for attacks, get unauthorized access and normal traffic, instead of predicting the type of specific attacks. A latest dataset is used to comprehensively evaluate the performance of proposed methods. And by comparing with the current attention based methods, our model shows state-of-the-art performance with overall accuracy of 93.63% across all proposed abstract attack intentions in independent validation. To conclude, we have proposed the idea of abstract attack intention inference, and the proposed interactive attention framework can effectively identify the abstract attack intentions that deriving from the internal devices of Internet of Things network, showcasing as a new attention based baseline model for Internet of Things tabular data based attack intention inference.