Low Communication Cost Research Articles

The application and challenges of different face recognition technologies in the three major fields of security, social media, and medical care

Abstract. Face recognition technology is a very important part of modern technology, which can not only be used to ensure security, but also can be used in the field of information organization and content division. However, with the popularization and application of face recognition technology, many problems that need to be solved urgently have emerged: excessive computing resources are consumed in order to pursue high-precision recognition, which brings computing pressure; The basis for improving recall is the need for a lot of power and memory; If security is not guaranteed, it can cause problems such as data breaches. The demand for face recognition technology is different in different use fields, so the purpose of this study is to combine the scene requirements and technical advantages more reasonably. The research results are as follows: the high accuracy and recall rate of 3D convolutional neural networks (3DCNNs) ensure that it can be used safely in high-precision and high-security scenarios. Lightweight Convolutional Neural Network (MobileNetV2) is suitable for resource-constrained environments due to its low memory consumption and low communication cost. Edge computing real-time face recognition (EC-RFERNet) is the most suitable for large-scale popularization and application among the three because of its lowest power consumption and latency. This study deeply explores the advantages and disadvantages of different facial recognition technologies and finds solutions to their shortcomings. According to their unique advantages combined with the requirements of commonly used scenarios, it provides a scientific basis for the deployment of face recognition technology in different fields. However, due to limited information, this paper cannot cover all application scenarios and the latest technologies, so it is hoped that in the future, we can combine the advantages of different technologies to develop more comprehensive face recognition technology and make more reasonable technical planning.

A privacy protection scheme to resist ring selection attack in peer-to-peer double auction V2V electricity transaction for charging stations

With the development of Electric Vehicles (EVs), the number of public charging piles cannot meet the immediate charging needs of EV users sometimes. The charging stations provide flexible electricity transaction schemes for users by managing charging piles and Vehicle-to-Vehicle (V2V). Users can freely choose transaction partners and submit transaction plans to multiple charging stations. The mobility of EVs results in users randomly selecting rings multiple times, which may cause them to appear in different rings. Attackers can link multiple transaction plans and infer user privacy information. The leakage of transaction information may threaten the fairness of transactions. Aiming at the above situations, a privacy protection scheme to resist ring selection attack in Peer-to-Peer double auction V2V electricity transaction for charging stations is proposed. Firstly, the ring signcryption algorithm is improved to resist ring selection attack. Secondly, a real identity verification algorithm is designed to track malicious users. Theoretical analysis proves that the security of the proposed privacy protection scheme. The elliptic curve point multiplication is used in the scheme to improve computational efficiency. Experiments show that this scheme has low computational and communication costs. Verification cost increases slowly with the increase of the number of ring members.

Achieving the Exactly Optimal Privacy-Utility Trade-Off With Low Communication Cost via Shared Randomness

Achieving the Exactly Optimal Privacy-Utility Trade-Off With Low Communication Cost via Shared Randomness

Lightweight ring-neighbor-based user authentication and group-key agreement for internet of drones

As mobile internet and Internet of Things technologies continue to advance, the application scenarios of peer-to-peer Internet of Drones (IoD) are becoming increasingly diverse. However, the development of IoD also faces significant challenges, such as security, privacy protection, and limited computing power, which require technological innovation to overcome. For group secure communication, it is necessary to provide two basic services, user authentication and group key agreement. Due to the limited storage of IoD devices, group key negotiation requires lightweight calculations, and conventional schemes cannot satisfy the requirements of group communication in the IoD. To this end, a new lightweight communication scheme based on ring neighbors is presented in this paper for IoD, which not only realizes the identity verification of user and group key negotiation, but also improves computational efficiency on each group member side. A detailed security analysis substantiates that the designed scheme is capable of withstanding attacks from both internal and external adversaries while satisfying all defined security requirements. More importantly, in our proposal, the computational cost on the user side remains unaffected by the variability of the number of members participating in group communication, as members communicate in a non-interactive manner through broadcasting. As a result, the protocol proposed in this article demonstrates lower computational and communication costs in comparison to other cryptographic schemes. Hence, this proposal presents a more appealing approach to lightweight group key agreement protocol with user authentication for application in the IoD.

CESA: Communication efficient secure aggregation scheme via sparse graph in federated learning

As a distributed learning paradigm, federated learning can be effectively applied to the decentralized system since it can resolve the “data island” problem. However, it is also vulnerable to serious privacy breaches. Although existing secure aggregation technique can address privacy concerns, they also incur significant additional computation and communication costs. To address these challenges, this paper offers a Communication Efficient Secure Aggregation scheme. Firstly, the central server uses the communication delay between terminals as the weight of the fully terminal-connected graph to transform it into a sparse connected graph based on the minimal spanning tree. Secondly, instead of relying on central server for key advertisement, the terminals advertise keys via a neighboring terminal forwarding approach based on sparsely graph. Thirdly, we propose using the central server for auxiliary advertising to address unexpected terminal dropout. Simultaneously, we theoretically demonstrate our scheme’s security and have lower computation and communication costs. Experiments show that CESA can reduce the running time by 28.2% without sacrificing security and model accuracy compared to conventional secure aggregation when there are 10 terminals in the system.

ATHENA-FL: Avoiding Statistical Heterogeneity with One-versus-All in Federated Learning

Federated learning (FL) is a distributed approach to train machine learning models without disclosing private data from participating clients to a central server. Nevertheless, FL training struggles to converge when clients have distinct data distributions, which leads to an increased training time and model prediction error. We propose ATHENA-FL, a federated learning system that considers clients with heterogeneous data distributions to generate accurate models in fewer training epochs than state-of-the-art approaches. ATHENA-FL reduces communication costs, providing an additional positive aspect for resource-constrained scenarios. ATHENA-FL mitigates data heterogeneity by introducing a preliminary step before training that clusters clients with similar data distribution. To handle that, we use the weights of a locally trained neural network used as a probe. The proposed system also uses the one-versus-all model to train one binary detector for each class in the cluster. Thus, clients can compose complex models combining multiple detectors. These detectors are shared with all participants through the system's database. We evaluate the clustering procedure using different layers from the neural network and verify that the last layer is sufficient to cluster the clients efficiently. The experiments show that using the last layer as input for the clustering algorithm transmits 99.68% fewer bytes to generate clusters compared to using all the neural network weights. Finally, our results show that ATHENA-FL correctly identifies samples, achieving up to 10.9% higher accuracy than traditional training. Furthermore, ATHENA-FL achieves lower training communication costs compared with MobileNet architecture, reducing the number of transmitted bytes between 25% and 97% across evaluated scenarios.

2FAKA-C/S: A Robust Two-Factor Authentication and Key Agreement Protocol for C/S Data Transmission in Federated Learning

As a hot technology trend, the federated learning (FL) cleverly combines data utilization and privacy protection by processing data locally on the client and only sharing model parameters with the server, embodying an efficient and secure collaborative learning model between clients and aggregated Servers. During the process of uploading parameters in FL models, there is susceptibility to unauthorized access threats, which can result in training data leakage. To ensure data security during transmission, the Authentication and Key Agreement (AKA) protocols are proposed to authenticate legitimate users and safeguard training data. However, existing AKA protocols for client–server (C/S) architecture show security deficiencies, such as lack of user anonymity and susceptibility to password guessing attacks. In this paper, we propose a robust 2FAKA-C/S protocol based on ECC and Hash-chain technology. Our security analysis shows that the proposed protocol ensures the session keys are semantically secure and can effectively resist various attacks. The performance analysis indicates that the proposed protocol achieves a total running time of 62.644 ms and requires only 800 bits of communication overhead, showing superior computational efficiency and lower communication costs compared to existing protocols. In conclusion, the proposed protocol securely protects the training parameters in a federated learning environment and provides a reliable guarantee for data transmission.

HCFAIUN: A novel hyperelliptic curve and fuzzy extractor-based authentication for secure data transmission in IoT-based UAV networks

IoT-based UAV networks comprise interconnected UAVs outfitted with sensors and microcontrollers to simplify data exchange in environments such as smart cities. In light of open-access communication landscapes, IoT-based UAV networks could pose security challenges, encompassing authentication vulnerabilities and the inadvertent disclosure of location and other confidential information to unauthorised parties. Henceforth, we have proposed a lightweight and secure authentication protocol: Hyperelliptic Curve and Fuzzy Extractor based Authentication in IoT-based UAV networks (HCFAIUN) leveraging Hyperelliptic Curve Cryptography(HCC), Fuzzy Extractor (FE), XOR operations and hash functions. HCC's maximum key size is 80 bits, differing from the 160-bit requirement of the elliptic curve, making it apt for UAVs with limited resources. The proposed scheme utilises biometrics traits of users to avoid exposing data from stealing smart devices using FE. This protocol facilitates the mutual authentication of users and UAVs, allowing them to exchange a session key for secure communication. The Hyperelliptic Curve (HC) scalar multiplication protects the user's private key from attackers, even in public channels. The obfuscation identity of the user and UAVs generated through the hash function and timestamp makes the external user and UAV anonymous. The efficacy of this proposed framework is examined using the Scyther verification tool and Random oracle model-based formal analysis, and informal analysis is also discussed, which validates its robustness against well-known potential physical and logical attacks. The performance analysis shows that the HCFAIUN scheme has lower computation, communication, and storage costs, i.e., 3.832 ms and 1456 bits and 1128 bits, respectively, compared to existing schemes.

A secure and efficient authentication key agreement scheme for industrial internet of things based on edge computing

Industrial Internet of Things (IIoT) is a pivotal driving force behind the intelligent transformation of the global industrial system, bringing about a gradual shift in production methods. However, due to users completing remote access and data transmission through open channels, ensuring user legitimacy and data security has become an important factor in IIoT. In addition, resource limited users or devices need to collect and publish corresponding messages while also timely processing instant messages obtained from the network. Leveraging edge computing technology effectively mitigates the transmission time of messages, addressing the challenges of high computing pressure and prolonged response time associated with cloud server computing. Therefore, investigating authentication key agreement based on edge computing in the IIoT environment holds substantial theoretical significance and practical importance. This article introduces an IIoT authentication that ensures both security and efficiency by integrating elliptic curve cryptography and three-factor authentication. Within this scheme, the user and the edge server accomplish security authentication. Through security analysis, the proposed scheme is proven to meet the essential security requirements. Performance analysis shows that it achieves higher security, supports more functions, and achieves lower computational and communication costs compared to related schemes.

Communication cost-aware client selection in online federated learning: A Lyapunov approach

The proliferation of intelligence services brings data breaches and privacy infringement concerns. To preserve data privacy when training machine learning models, the federated learning (FL) paradigm emerges. Most existing works assume that training data on FL clients are static during the entire learning process. Nevertheless, various real-time intelligent services call for timely processing of continuously generated data, which fosters the advent of online federated learning (OFL). Currently, how to reconcile model utility and communication cost in OFL is still an open problem. To address this challenge, we leverage the Lyapunov optimization framework to devise a novel Low Cost Client Selection (LCCS) algorithm for OFL, which can judiciously select participating clients to maximize model utility with a low communication cost. Specifically, we design the objective as the sum of a penalty function and a Lyapunov drift function to take both gradient-based client valuation and communication cost into account. By minimizing the objective, we further design the LCCS algorithm, which is lightweight for execution on clients. At last, we conduct extensive experiments with traces generated from public datasets. The experimental results demonstrate that LCCS achieves the highest model utility with a fixed communication cost in comparison with the state-of-the-art baselines.

A PUF and Fuzzy Extractor-Based UAV-Ground Station and UAV-UAV Authentication Mechanism With Intelligent Adaptation of Secure Sessions

It is crucial that communication between an unmanned aerial vehicle (UAV) and the ground station (GS) be secure, and both devices should mutually authenticate each other to ensure that an adversary cannot obtain communicated information. Moreover, dynamically adapting the session time of an authenticated session can decrease the idle time of a session and consequently reduce the window of opportunity for an adversary to interfere with the communication link. In light of these considerations, we design a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">physically unclonable function (PUF)</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">fuzzy extractor</i> -based UAV-GS authentication mechanism called <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">UAV Authentication with Adaptive Session (UAAS)</i> . In UAAS, both UAV-GS and UAV-UAV authentication are two-way. We use a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Thompson Sampling (TS)</i> -based approach to intelligently adapt the duration of a session. Both formal and informal security proofs are presented along with a computation and communication cost analysis to analyze the performance of UAAS. It is noted that UAAS is secure against various well-known attacks and has a lower communication cost than several baseline mechanisms. Due to the noise reduction that occurs in PUFs, the computational cost of UAAS is higher than that of baselines that ignore noise. Also, our simulation shows that UAAS significantly outperforms baselines when it comes to network performance.

Secure data collection and transmission for IoMT architecture integrated with federated learning

PurposeThe purpose of this paper is to secure health data collection and transmission (SHDCT). In this system, a native network consists of portable smart devices that interact with multiple gateways. It entails IoMT devices and wearables connecting to exchange sensitive data with a sensor node which performs the aggeration process and then communicates the data using a Fog server. If the aggregator sensor loses the connection from the Fog server, it will be unable to submit data directly to the Fog server. The node transmits encrypted information with a neighboring sensor and sends it to the Fog server integrated with federated learning, which encrypts data to the existing data. The fog server performs the operations on the measured data, and the values are stored in the local storage area and later it is updated to the cloud server.Design/methodology/approachSHDCT uses an Internet-of-things (IoT)-based monitoring network, making it possible for smart devices to connect and interact with each other. The main purpose of the monitoring network has been in the collection of biological data and additional information from mobile devices to the patients. The monitoring network is composed of three different types of smart devices that is at the heart of the IoT.FindingsIt has been addressed in this work how to design an architecture for safe data aggregation in heterogeneous IoT-federated learning-enabled wireless sensor networks (WSNs), which makes use of basic encoding and data aggregation methods to achieve this. The authors suggest that the small gateway node (SGN) captures all of the sensed data from the SD and uses a simple, lightweight encoding scheme and cryptographic techniques to convey the data to the gateway node (GWN). The GWN gets all of the medical data from SGN and ensures that the data is accurate and up to date. If the data obtained is trustworthy, then the medical data should be aggregated and sent to the Fog server for further processing. The Java programming language simulates and analyzes the proposed SHDCT model for deployment and message initiation. When comparing the SHDCT scheme to the SPPDA and electrohydrodynamic atomisation (EHDA) schemes, the results show that the SHDCT method performs significantly better. When compared with the SPPDA and EHDA schemes, the suggested SHDCT plan necessitates a lower communication cost. In comparison to EHDA and SPPDA, SHDCT achieves 4.72% and 13.59% less, respectively. When compared to other transmission techniques, SHDCT has a higher transmission ratio. When compared with EHDA and SPPDA, SHDCT achieves 8.47% and 24.41% higher transmission ratios, respectively. When compared with other ways it uses less electricity. When compared with EHDA and SPPDA, SHDCT achieves 5.85% and 18.86% greater residual energy, respectively.Originality/valueIn the health care sector, a series of interconnected medical devices collect data using IoT networks in the health care domain. Preventive, predictive, personalized and participatory care is becoming increasingly popular in the health care sector. Safe data collection and transfer to a centralized server is a challenging scenario. This study presents a mechanism for SHDCT. The mechanism consists of Smart healthcare IoT devices working on federated learning that link up with one another to exchange health data. Health data is sensitive and needs to be exchanged securely and efficiently. In the mechanism, the sensing devices send data to a SGN. This SGN uses a lightweight encoding scheme and performs cryptography techniques to communicate the data with the GWN. The GWN gets all the health data from the SGN and makes it possible to confirm that the data is validated. If the received data is reliable, then aggregate the medical data and transmit it to the Fog server for further process. The performance parameters are compared with the other systems in terms of communication costs, transmission ratio and energy use.

IoV-6G+: A secure blockchain-based data collection and sharing framework for Internet of vehicles in 6G-assisted environment

The growing need for wide and ubiquitous accessibility to advanced Intelligent Transportation Systems (ITS) has led to the evolution of conventional Vehicle to Everything (V2X) paradigms into the Internet of Vehicles (IoVs). Next-generation IoVs establish seamless connections among humans, vehicles, Internet of Things (IoT) devices, and service platforms to enhance transit efficiency, road safety, and environmental conservation, with notable advancements in IoV technologies, specifically in secure data exchange and user privacy protection. Additionally, Unmanned Aerial Vehicles (UAVs) are envisioned as scalable and adaptable solutions for comprehensive IoV service coverage. The current approaches in IoV mainly concentrate on local infrastructure setups, overlooking the potential of infrastructure-less IoVs that lack extensive edge facilities. Also, the transmission of data over public channels by vehicles equipped with servers is susceptible to interception and tampering by malicious attackers. Moreover, the substantial amount of real-time data generated by vehicles, IoT devices, travelers, and social interactions pose a significant strain on servers as well as latency issues. To address these security concerns, this research suggests a secure-lightweight data collection and sharing framework for 6G-assisted smart transportation using blockchain and UAV, called IoV-6G+. The process entails drones collecting information and transmitting it to dedicated edge servers that securely aggregate the information and generate transactions on a cloud server. A cloud server securely collects data from edge servers, creates transactions, combines them into blocks, and then confirms and adds these blocks to the blockchain through a voting-based consensus method in a peer-to-peer network of cloud servers. Additionally, the security analysis of our proposed IoV-6G+ is assessed using the “Informal and formal (i.e., Scyther Tool, and Real-or-Random (RoR) Model) methods”, showcasing its efficacy in delivering a secure and efficient authentication mechanism for IoV. Moreover, the proposed IoV-6G+ exhibits lower computational and communication costs as well as energy consumption, along with enhanced security features, when compared to existing authentication frameworks.

A lattice-based efficient certificateless public key encryption for big data security in clouds

With the advancement of technology and the development of digitization, big data has become an integral component of modern society. Due to the huge volume of big data, users generally upload data to the cloud for storage and computation. Because data in the cloud is beyond the control of users, it faces threats such as data security and personal privacy leakage. To address these concerns and achieve big data security and privacy protection in clouds, data encryption is crucial. Traditional public key encryption involves complex certificate management in a public key infrastructure (PKI). Although the identity-based encryption (IBE) scheme avoids the management of public key certificates, it has issues with key escrows and identity revocation. The certificateless public key encryption (CL-PKE) not only solves the key escrows problem in IBE, but also maintains its advantage of not requiring the use of certificate in PKI, making it suitable for the security protection of big data in clouds. However, most CL-PKE schemes are constructed based on large integer factorization and discrete logarithm problems, which cannot resist quantum computing attacks. Therefore, this paper proposes an efficient lattice-based CL-PKE scheme for big data security in clouds. In addition, the proposed scheme is extended to a lattice-based certificateless proxy re-encryption scheme, facilitating secure big data sharing in clouds. The proposed CL-PKE scheme can resist attacks from both internal and external adversaries, and has been proven to be IND-CPA secure under the learning with errors (LWE) assumption. Experiments have shown that the proposed CL-PKE scheme has lower computational and communication costs. The proposed lattice-based schemes can effectively provide post quantum security for the security and privacy protection of big data in clouds.

Multipars: Reduced-Communication MPC over Z2k

In recent years, actively secure SPDZ-like protocols for dishonest majority, like SPDZ2k, Overdrive2k, and MHz2k, over base rings Z2k have become more and more efficient. In this paper, we present a new actively secure MPC protocol Multipars that outperforms these state-of-the-art protocols over Z2k by more than a factor of 2 in the two-party setup in terms of communication. Multipars is the first actively secure N-party protocol over Z2k that is based on linear homomorphic encryption (LHE) in the offline phase (instead of oblivious transfer or somewhat homomorphic encryption in previous works). The strong performance of Multipars relies on a new adaptive packing for BGV ciphertexts that allows us to reduce the parameter size of the encryption scheme and the overall communication cost. Additionally, we use modulus switching for further size reduction, a new type of enhanced CPA security over Z2k, a truncation protocol for Beaver triples, and a new LHE-based offline protocol without sacrificing over Z2k. We have implemented Multipars and therewith provide the fastest preprocessing phase over Z2k. Our evaluation shows that Multipars offers at least a factor of 8 lower communication costs and up to a factor of 15 faster runtime in the WAN setting compared to the currently best available actively secure MPC implementation over Z2k.

Exploiting Label Skews in Federated Learning with Model Concatenation

Federated Learning (FL) has emerged as a promising solution to perform deep learning on different data owners without exchanging raw data. However, non-IID data has been a key challenge in FL, which could significantly degrade the accuracy of the final model. Among different non-IID types, label skews have been challenging and common in image classification and other tasks. Instead of averaging the local models in most previous studies, we propose FedConcat, a simple and effective approach that concatenates these local models as the base of the global model to effectively aggregate the local knowledge. To reduce the size of the global model, we adopt the clustering technique to group the clients by their label distributions and collaboratively train a model inside each cluster. We theoretically analyze the advantage of concatenation over averaging by analyzing the information bottleneck of deep neural networks. Experimental results demonstrate that FedConcat achieves significantly higher accuracy than previous state-of-the-art FL methods in various heterogeneous label skew distribution settings and meanwhile has lower communication costs. Our code is publicly available at https://github.com/sjtudyq/FedConcat.

Improvement of a Conditional Privacy-Preserving and Desynchronization-Resistant Authentication Protocol for IoV

In Internet of Vehicles (IoV), the secure data transmission between vehicles and transportation infrastructure effectively ensures the safety and fast driving of vehicles, while authentication and key agreement protocols between vehicles and transportation infrastructure (V2I) play an important role in ensuring data security and user privacy. However, how to design a provably secure and lightweight V2I authentication protocol for IoV is a challenge. Recently, Kumar et al. proposed a conditional privacy-preserving and desynchronization-resistant authentication protocol for IoV, but we find that Kumar et al.’s scheme is vulnerable to identity guessing attacks, impersonation attacks, and a lack of session key secrecy once the attacker obtains data stored in smart card through side-channel attacks. We also point out that Kumar et al.’s protocol is vulnerable to Roadside Unit (RSU) captured attacks and lacks perfect forward secrecy. Therefore, we propose an improved V2I authentication protocol for IoV, which uses the Physical Unclonable Function (PUF) to resist RSU captured attacks, and designed a three-factor secrecy strategy to resist side-channel attacks; a conditional privacy-preserving strategy was also adopted to achieve anonymity and malicious user tracking. Furthermore, the proposed protocol is provably secure under the random oracle model and has low computation and communication costs.

A Lightweight Chaotic Map-Based Key Agreement Scheme for the Advanced Metering Infrastructure

In the advanced metering infrastructure (AMI), impersonation, eavesdropping, man-in-the-middle and other attacks occur in the process of communication between entities through public channels, which will lead to the leakage of user privacy or the incorrect issuance of control instructions, resulting in economic losses and even power system operation failures. In view of this situation, we design a lightweight key agreement scheme based on a chaotic map for the AMI. We use the chaotic map to replace the time-consuming bilinear pairing and elliptic curve method and establish a secure communication channel between legal entities. In addition, we also design a multicast key generation mechanism for message transmission in AMI. The security analysis proves the security of the proposed scheme in the random oracle model, which can meet the security characteristics of anonymity and forward secrecy, and can effectively resist common attacks such as impersonation, replay and man-in-the-middle. The performance analysis results show that the proposed scheme requires lower computational and communication costs than related schemes, so it is more suitable for AMI scenarios with limited resources.

DAC‐MD: A privacy preserving drone‐access control scheme for last mile delivery

AbstractThe service delivery has lately witnessed strides in the form of paradigm shifts from conventional logistics to drone‐oriented supply chain to conserve ecosystem. The use of drones or unmanned aerial vehicles (UAVs) for package deliveries offer enhanced accessibility, flexibility, efficiency, speed, safety, and even few environmental gains. However, the drones can come across various security hazards while communicating with ground installations on open channels including physical capture attack, privacy infringement, and so on. Thus, earnest endeavors are needed to fix those security challenges in applications related to drone communications. To address these problems, an efficient authenticated key agreement technique, drone access control scheme for last mile delivery (DAC‐MD) is suggested for user‐drone interactions in IoD environment that help the participating entities in developing a mutually agreed session key for securing subsequent communication interactions. The proposed scheme is proved using automated ProVerif tool as well as analyzed by using formal analysis based on RoR random oracle model. The comparative analysis against recently presented known studies reveals that our scheme not only carries comparable properties in terms of security and privacy, but also low computational and communication costs.

Dynamic ensemble of regression neural networks based on predictive uncertainty

In a data-decentralized environment where data are distributed across individual local nodes, inter-node data sharing is often restricted due to security concerns and communication costs. Under these restrictions, using entire data to train a global predictive model is challenging. A viable strategy is to leverage local predictive models trained independently on local nodes. However, this adversely affect generalization performance, particularly when the data distribution varies between local nodes. To address this issue in regression problems, we propose a dynamic ensemble method that combines the predictions of locally trained neural networks without the need to access data from local nodes. For a query instance, the prediction and its uncertainty from each neural network are computed by applying Monte Carlo dropout. Subsequently, smaller/larger weights are assigned to neural networks with higher/lower predictive uncertainty. The final prediction result is derived using the weighted average of the individual predictions. The proposed method is applicable to a data-decentralized environment where data are inaccessible and thus cannot be shared across local nodes. Transmitting query instances to local nodes and receiving predictions and weights from the local nodes are the only requirements, thereby ensuring low communication costs. Experimental results on simulated data-decentralized environments using benchmark regression datasets validated the superior generalization performance of the proposed method compared with baseline methods. The proposed method was found to be more effective for regression neural networks that demonstrate higher performance at their respective local nodes.