To cloud or not to cloud: an on-line scheduler for dynamic privacy-protection of deep learning workload on edge devices

Abstract

Recently deep learning applications are thriving on edge and mobile computing scenarios, due to the concerns of latency constraints, data security and privacy, and other considerations. However, because of the limitation of power delivery, battery lifetime and computation resource, offering real-time neural network inference ability has to resort to the specialized energy-efficient architecture, and sometimes the coordination between the edge devices and the powerful cloud or fog facilities. This work investigates a realistic scenario when an on-line scheduler is needed to meet the requirement of latency even when the edge computing resources and communication speed are dynamically fluctuating, while protecting the privacy of users as well. It also leverages the approximate computing feature of neural networks and actively trade-off excessive neural network propagation paths for latency guarantee even when local resource provision is unstable. Combining neural network approximation and dynamic scheduling, the real-time deep learning system could adapt to different requirements of latency/ accuracy and the resource fluctuation of mobile-cloud applications. Experimental results also demonstrate that the proposed scheduler significantly improves the energy efficiency of real-time neural networks on edge devices.

An Enhanced Intrusion Detection System for IoT Networks Based on Deep Learning and Knowledge Graph

“…) end for (5) for each f i ∈ F do (6) Calculate pairwise feature pairs and judge whether the features cooccur. ( 7) NumPairs(〈f i , f j 〉) � n ij + 0, if there is no cooccurrence relationship n ij + 1, if features f i and f j are cooccurring 􏼨 (8) end for (9) for each 〈f i , f j 〉 from NumPairs do (10) analyzed in intrusion detection, it is difficult to judge whether it is a port scan. However, when multiple data packets are serialized and the LSTM network is used for judgment, it can be more accurately judged that these data packets are from port scanning attacks, with the reason that the LSTM network can learn context information and serialized features.…”

“…(3) for each 〈f i , f j 〉 ∈ M do (4) F KG � FeaturesOfKG.append(f i , f j ) (5) F KG � AlignmentBySort(F KG ) (6) end for (7) Fuse the features extracted from the two views of KG and SA. (8) U � F KG ⋃ F SA (9) for each f i ∈ F do (10) Weight calculation by relationship between the number of host connections and a specific flag feature of the connection rejection error (e.g., REJ). Obviously, this processing can maintain contextual semantic information, and the BiLSTM is a coarse-grained intrusion detection model.…”

“…In (10), h t → and h t ← , respectively, represent the state of the forward LSTM layer and the backward LSTM layer at time t, corresponding to the context feature of IoTnetwork traffic. s t is the word vector input at time t, and w 1 to w 6 represent the layer's weight parameters.…”

“…In this paper, we focus on constructing IDS for IoT networks. Meanwhile, machine learning (ML) has been widely used in security research recently, such as APT attack detection [9], personal privacy protection [10], malicious code analysis [11], cyberphysical system defense [8], and malicious traffic detection [12]. In order to detect malicious behaviors in large-scale network traffic of IoT, machine learning-based IDS has attracted widespread attention.…”