10 December 2021 in IEEE Transactions on Artificial Intelligence
Wider coverage and a better solution to a latency reduction in 5G necessitate its combination with multi-access edge computing technology. Decentralized deep learning (DDL), such as federated learning and swarm learning, as a promising solution to privacy-preserving data processing for millions of smart edge devices leverages distributed computing of multilayer neural networks within the networking of local clients, without disclosing the original local training data. Notably, in industries such as finance and healthcare, where sensitive data of transactions and personal medical records are cautiously maintained, DDL can facilitate the collaboration among these institutes to improve the performance of trained models while protecting the data privacy of participating clients. In this survey article, we demonstrate the technical fundamentals of DDL that benefit many walks of society through decentralized learning. Furthermore, we offer a comprehensive overview of the current state of the art in the field by outlining the challenges of DDL and the most relevant solutions from novel perspectives of communication efficiency and trustworthiness.
14 February 2024 in IEEE Access
Electricity theft is a pervasive issue with economic implications that necessitate innovative approaches for its detection, given the critical challenge of limited labeled data. However, connecting smart home devices introduces numerous vectors for electricity theft. Therefore, this study introduces an innovative approach to detecting electricity theft in smart homes, leveraging the knowledge-based fine-grained time-series appliance benign and anomalous consumption patterns. We simulated five attack classes and extended our model’s detection capabilities to unknown anomalies across residential settings by segmenting the anonymized data into three different home categories. We validated our experiment using simulated and actual building attack data. The synthetic binary discriminator model (SYNBDM)—Extreme Gradient Boost (XGB), Random Forest, and Multilayer Perceptron (MLP)—outperformed the legacy unsupervised model (LUM)—MLP-Autoencoder(AE), 1D-CONV-AE, and Isolation Forest (RF). XGB had the highest average AUCscores of 98.69% and 98.74% for simulated and attack detection respectively, followed by RF at 96.76% and 97.07%, respectively, across all homes, indicating the robustness of our model in detecting benign and anomalous appliance consumption patterns. This study contributes to the academic discourse in the field and offers practical solutions to energy providers and stakeholders in the smart home industry.
11 April 2022 in Expert Systems with Applications
20 December 2021 in IEEE Sensors Journal
Maneuvering target tracking in wireless sensor networks (WSNs) has been a prominent research topic over the past twenty years. In many situations, WSNs are vulnerable to natural or man-made interference, resulting in uncertainty problems. In this paper, we propose an information-quality-based sensor selection method with estimation feedback for maneuvering target tracking in uncertain WSNs. First, based on the Generalized Pseudo-Bayesian estimator of first order (GPB1) and the unscented information filter (UIF), a GPB1-UIF algorithm with estimation feedback is presented. Inspired by the theory of information quality, an information-quality metric framework is proposed for sensor selection in uncertain WSNs. And the information-theoretic selection metrics are redesigned based on consistency measuring operator. Moreover, without knowing the number of selected sensors a priori, a multi-objective optimization problem is considered to determine the final selection strategy. Simulation results demonstrate that the proposed approach outperform the previous sensor selection methods.
26 May 2023 in IEEE Transactions on Artificial Intelligence
Existing model poisoning attacks on federated learning (FL) assume that an adversary has access to the full data distribution. In reality, an adversary usually has limited prior knowledge about clients' data. A poorly chosen target class renders an attack less effective. This work considers a semi-targeted situation where the source class is predetermined but the target class is not. The goal is to cause the misclassification of the global classifier on data from the source class. Approaches such as label flipping have been used to inject malicious parameters into FL. Nevertheless, it has shown that their performances are usually class-sensitive, varying with different target classes. Typically, an attack becomes less effective when shifting to a different target class. To overcome this challenge, we propose the Attacking Distance-aware Attack (ADA) that enhances model poisoning in FL by finding the optimized target class in the feature space. ADA deduces pair-wise class attacking distances using a Fast LAyer gradient MEthod (FLAME). Extensive evaluations were performed on five benchmark image classification tasks and three model architectures using varying attacking frequencies. Furthermore, ADA's robustness to conventional defenses of Byzantine-robust aggregation and differential privacy was validated. The results showed that ADA succeeded in increasing attack performance to 2.8 times in the most challenging case with an attacking frequency of 0.01 and bypassed existing defenses, where differential privacy that was the most effective defense still could not reduce the attack performance to below 50%.
21 June 2022 in Scientific Data
This paper reports the release of dataset that describes the critical city inbound-outbound intersection in the central business district’s Sathorn area of Bangkok, Thailand. The available traffic volume and occupancy are captured by sensors located on the Sathorn area’s links. Induction loop coil sensors and thermal and CCTV cameras have been installed at the approaching links of the critical Sathorn-Surasak intersection. Traffic volume data have been collected from the CCTV cameras every 5 s from 2016 to 2019, and occupancy and volume data have been collected from the loop coil sensors every 5 s from May to September 2016 during a social experiment that was part of Sathorn Model project. Occupancy and volume data have also been collected from the thermal cameras every 5 s from May to June 2016. The dataset provides temporal and spatial coverage of Sathorn Road’s primary urban areas, including weekdays, weekends, and public holidays. This dataset can be resources for research on traffic state estimation, traffic light control optimization, and the analysis of critical intersections.