Parameter-efficient fine-tuning (PEFT) has become the mainstream paradigm for adapting large language models (LLMs) to downstream tasks. However, existing PEFT approaches often suffer performance degradation when applied to complex tasks. To address this limitation, we propose a hierarchical Pyramid-Structured Low-Rank Adaptation (PRLoRA) method. PRLoRA constructs a pyramid network architecture characterized by multi-scale representations that effectively balance local precision with global contextual awareness. In addition, it incorporates a local optimization module that evaluates the importance of LLM weights and adaptively selects locally optimal positions within the pyramid, thereby accelerating convergence. The pyramid structure fusion LoRA (PSFLoRA) module within PRLoRA fuses multi-scale features of the pyramid to enhance the method’s effective rank. We conduct comprehensive experiments on 14 datasets; PRLoRA achieves state-of-the-art performance on both the GLUE benchmark and arithmetic reasoning tasks. In particular, it outperforms LoRA by 6.33% on AddSub and surpasses the previous state-of-the-art, DoRA, by 1.89% on GSM8K. Extensive ablation studies validate the contribution of each component and demonstrate PRLoRA’s robustness across diverse architectures (LLaMA3, OPT, BLOOM, Gemma) and multiple model sizes (1B, 3B, 8B).

Decentralized federated learning (DFL) has been widely used in edge computing and Internet of Things (IoT) settings with many devices. However, the heterogeneity of devices (e.g., varying computational capacity, stability, security requirements) can impact the performance of DFL applications. Our proposed SSAA, a secure aggregation scheme for DFL on heterogeneous devices, presented in this paper is designed to improve the efficiency of DFL while preserving privacy. Specifically, SSAA accelerates aggregation by synchronously coupling aggregation device-set formation with aggregation computation, and can cope with device availability and performance variability to maintain stable and efficient aggregation. By extending homomorphic encryption to support cross-round ciphertext continuity, SSAA enables reliable and secure decryption under large-scale dropouts in the original aggregation set, making it practical for dynamic DFL environments. In addition, we prove that SSAA is semi-honestly secure and resistant to device collusion attacks – fundamental security requirements for applications involving heterogeneous devices. We also implement SSAA and comprehensively evaluate its performance to demonstrate its practicability.

While multi-key homomorphic encryption (MKHE) ensures privacy in federated learning (FL) by encrypting model updates, its requirement for aggregate ciphertext decryption and dropout handling increases communication rounds during aggregation. To enable secure multi-key aggregation with low communication rounds, we propose SAMK. SAMK enables the server to compute the sum of model updates from clients participating in FL, while keeping these updates encrypted by different keys throughout the computation. By utilizing the polynomial property of the BFV ciphertext, SAMK successfully implements individual decryption of the aggregated ciphertext (encrypted under multiple keys) by each client using their respective keys, resulting in low communication rounds. It means that in SAMK, all client interactions are avoided and the client-server interaction is only once (ciphertext uploads and downloads) in each round of aggregation computation. In addition, SAMK is robust to any number of clients dropping out at any time, and the client who has dropped out after uploading model updates, can still get the correct aggregation result upon reconnecting. We prove the security of SAMK for semi-honest server and clients, where client collusion is also considered. At last, we implement SAMK and comprehensively evaluate its performance to demonstrate its practicability.

Federated learning (FL), as a distributed machine learning paradigm, enables multiple users to train machine learning models locally using individual data and then update global model in a privacy-preserving aggregated manner. However, in FL, the users model parameters are at risk of a privacy breach. Furthermore, the aggregation server may forge aggregated results. To address these problems, in this paper, we propose SVCA, a secure and verifiable chained aggregation for privacy-preserving federated learning (PPFL) scheme. Specifically, we first group users and construct a chained aggregation structure, then employ secret sharing to prevent the entire group of users dropout, and finally propose a scheme for secure verification of the aggregation result to ensure the result correctness and the security of the verification process. The security analysis shows that SVCA not only protects the privacy of users but also ensures the training integrity. Extensive experimental results demonstrate the practical performance of SVCA without compromising classification accuracy.

Dynamic searchable symmetric encryption (DSSE) allows efficient searches over encrypted databases and also supports clients in their updating of the data, such as those stored in a remote cloud server. However, recent attacks suggest the risk of leakage during such updates, which consequently impacts on the privacy of the queries. In addition, existing DSSE schemes that support forward privacy generally rely on the honest-but-curious server and support only single-keyword retrieval, which limits the application scenarios. In this paper, we present the design of a verifiable DSSE protocol, which supports efficient conjunctive query with forward privacy. In our scheme, the forward index is constructed by a novel form, i.e.,$ t$-puncturable PRFs, and the authentication tag is designed by symmetric cryptography. During conjunctive queries, we narrow the scope by an inverted index, and then we determine the results of the final query through the forward index. Meanwhile, we can use verification tag to check the correctness and completeness of the result. In addition, we present an extension to support backward privacy, and our experimental evaluations show that our proposed approach achieves better performance on both conjunctive queries and updates than other competing solutions and ensures efficient verification.

Federated learning (FL) has been used to enhance privacy protection in edge computing systems. However, attacks on uploaded model gradients may lead to private data leakage, and edge devices frequently joining and leaving will impact the system running. In this paper, we propose a dynamic and flexible federated edge learning (FEL) scheme that can defend against malicious edge servers and edge devices to recover sensitive data and efficiently manage edge devices. A heterogeneity-aware scheduling strategy is designed to take into account the different impacts of heterogeneous edge devices on global model performance. The strategy determines the order of devices participation in each round based on the relative contribution level of the online edge device model, and the edge device with the highest contribution level is selected first. Numerical experiments show that our system improves test accuracy and time, and the security analyses show that our scheme meets the security requirements.

Electronic wearable devices play an important role in the Internet of Things (IoT) systems for collecting medical data. Searching over such numerical medical data help to provide better service and treatment. However, user and data security is a major barrier to public adoption. Existing approaches designed to facilitate secure (range) searches over encrypted data generally incur expensive computational overhead suffer from unexpected information leakage, and/or have high false-positive results. Therefore, in this article, we design a hybrid searchable encryption scheme that supports efficient, secure, and accurate range searches over encrypted data sensed and collected from medical IoT devices. The designed graph structure helps to filter out most of the false data, and the batching processing on ciphertexts accelerates the removal of irrelevant data. Unlike most prior works, the proposed random index hides the distribution of data, and the probabilistic fixed-length trapdoor hides the range size and repetition of the query. If necessary, all the encrypted data can be refreshed by the cloud server after a range search. The scheme is proven to be secure in a simulation-based model. Then, we evaluate the performance of our proposed scheme on Microsoft Azure cloud servers and Azure IoT Central. The comparisons with several prior works demonstrate that our scheme supports more efficient secure range searches.

The outsourcing of data is becoming increasingly commonplace as data is constantly been synchronized between user systems (e.g., personal computers and sensor devices) and cloud computing servers. However, to ensure data privacy, it is necessary to encrypt sensitive data prior to outsourcing. Limitations such as measurement, network delays, and data obfuscation may, however, result in uncertain data. Compared with searching over encrypted certain data, processing queries for encrypted uncertain data is more challenging. In this article, we propose a secure and efficient range query scheme over outsourced encrypted uncertain data, for example, from Internet of Things (IoT) systems. Specifically, we use pivot mapping to map data to a low-dimensional space to facilitate calculation and processing while preserving some of the original relevance among the data. Additionally, we encode data and then map codes into multiple Bloom filters which are organized by a binary tree-based index. Our scheme achieves data privacy, hides the relevance among data, and also supports efficient queries. We analyze the security and evaluate the performance of our approach using experiments on Microsoft Azure. The analysis and experimental results demonstrate that our proposed approach is secure and efficient.

Searchable symmetric encryption (SSE) enables a user to outsource a private dataset to a cloud server in encrypted form while retaining the ability to search over the encrypted outsourced data. The existing SSE schemes improve the search and safety performances from different perspectives. However, almost none of the existing SSE schemes considers the data distribution issues. We find that when the dataset is not distributed uniformly, the search quality based on the conventional methods decreases. Therefore, the existing SSE schemes cannot guarantee high search quality when faced with non-uniform datasets. In addition, most existing SSE solutions cannot hide the distribution of the query set. In this article, we design a S\ecure similarity search over Encrypted Non-uniform and high-dimensional Datasets (SEND) with a novel way to enhance security. The basic idea is to combine SSE with locality-sensitive hashing (LSH). Unlike earlier schemes, SEND uses selective hashing, which has better performance for non-uniform datasets. Also, we present a novel approach to hide the distribution of the query set, which makes SEND more secure. Our experimental results indicate SEND achieves a high search quality of recall and precision, and it is proven secure against adaptively chosen query attacks in the standard model.

There exist two severe challenges in cloud file-sharing networks: cooperation dilemma and trust dilemma. The mechanism designed to promote cooperation could suffer from malicious users, while the trust management that only considers the trust dilemma is subjected to denial-of-service attacks. To address these two dilemmas simultaneously, we present a dependable cloud file-sharing scheme-EigenCloud. The main contributions include the following. First, we propose a modified EigenTrust algorithm to calculate the global cooperation value and global trust value of each cloud user based on her/his past behaviors. Second, we propose cooperation and trust-aware worker recommendation mechanism by determining a Pareto front from all cloud users. Thus, a cloud user who adopts the recommendation mechanism by paying an additional fee could have a higher probability of receiving a valid file in one transaction. Last but not least, we use the evolutionary game theory (EGT) to study the acceptance and effectiveness of the proposed EigenCloud by strategically modeling cloud users. The Lyapunov stability theory is employed to mathematically investigate the stability of evolutionary equilibriums of our EigenCloud. Finally, both numerical simulations and simulator-driving experiments illustrate that our EigenCloud has an outstanding performance in promoting cooperation and inhibiting malicious activity.

Medical imaging is crucial for medical diagnosis, and the sensitive nature of medical images necessitates rigorous security and privacy solutions to be in place. In a cloud-based medical system for Healthcare Industry 4.0, medical images should be encrypted prior to being outsourced. However, processing queries over encrypted data without first executing the decryption operation is challenging and impractical at present. In this paper, we propose a secure and efficient scheme to find the exact nearest neighbor over encrypted medical images. Instead of calculating the Euclidean distance, we reject candidates by computing the lower bound of the Euclidean distance that is related to the mean and standard deviation of data. Unlike most existing schemes, our scheme can obtain the exact nearest neighbor rather than an approximate result. We, then, evaluate our proposed approach to demonstrate its utility.

Wearable body area network is a key component of the modern-day e-healthcare system (e.g., telemedicine), particularly as the number and types of wearable medical monitoring systems increase. The importance of such systems is reinforced in the current COVID-19 pandemic. In addition to the need for a secure collection of medical data, there is also a need to process data in real-time. In this article, we design an improved symmetric homomorphic cryptosystem and a fog-based communication architecture to support delay- or time-sensitive monitoring and other-related applications. Specifically, medical data can be analyzed at the fog servers in a secure manner. This will facilitate decision making, for example, allowing relevant stakeholders to detect and respond to emergency situations, based on real-time data analysis. We present two attack games to demonstrate that our approach is secure (i.e., chosen-plaintext attack resilience under the computational Diffie-Hellman assumption), and evaluate the complexity of its computations. A comparative summary of its performance and three other related approaches suggests that our approach enables privacy-assured medical data aggregation, and the simulation experiments using Microsoft Azure further demonstrate the utility of our scheme.

Significant volume of information of a broad variety (or modalities, such as image, audio, video, and text) is sensed and collected [such as those by the Internet of Things (IoT) devices] regularly (e.g., hourly). Such information is then analyzed to inform decision making, such as clinical diagnosis and product recommendation. Data with different representations may have the same semantic information, and there have been considerable efforts devoted to designing efficient searching approaches on objects with different modalities. However, multimodal data carry sensitive information, and maintaining privacy is crucial in our privacy-aware and interconnected society. In this article, we combine both the collective matrix factorization (CMF) and homomorphic encryption (HE) to construct an efficient and accurate scheme to facilitate cross-modal retrieval, without the loss of any sensitive information. Our scheme identifies the unified feature vectors for every object in the training set with different modalities and obtains the mapping matrices for out-of-sample objects. After the encryption process, these matrices are stored on the remote cloud server (CS). Hence, the server can calculate the secure, unified features for any query. In this article, we also built a privacy-preserving index structure using locality-sensitive hashing (LSH), which provides both security and efficiency. Performance evaluations demonstrate the potential for our proposed scheme in the real-world IoT applications.

The significant volume, variety, and velocity of data received from the many Industrial Internet of Things (IIoT) devices and other systems in a cloud-based or fog-based environment can complicate an organization's effort in ensuring high quality of experience for data users (DUs). For example, how do we efficiently and fairly allocate resources among cloud centers (CCs), fog service providers (FSPs), and DUs? This is particularly crucial for the IIoT environment, such as those in critical infrastructure sectors, such as energy and dams. Therefore, in this article, we propose an optimal resource allocation scheme for a fog-based IIoT environment. Specifically, we introduce fog nodes (or FSPs) that compete with each other to provide services for the DUs using resources from the CC. To maximize resource utilization, we model the resource allocation problem as a double-stage Stackelberg game and propose three algorithms to achieve Nash equilibrium and Stackelberg equilibrium. Then, we evaluate the performance of our proposed scheme with and without having FSPs, as well as with another competing scheme. The findings demonstrate the importance of fog computing in resource allocation, and the performance of our scheme outperforms that of the other scheme.

As cloud computing becomes prevalent, more and more data owners are likely to outsource their data to a cloud server. However, to ensure privacy, the data should be encrypted before outsourcing. Symmetric searchable encryption allows users to retrieve keyword over encrypted data without decrypting the data. Many existing schemes that are based on symmetric searchable encryption only support single keyword search, conjunctive keywords search, multiple keywords search, or single phrase search. However, some schemes, i.e., static schemes, only search one phrase in a query request. In this paper, we propose a multi-phrase ranked search over encrypted cloud data, which also supports dynamic update operations, such as adding or deleting files. We used an inverted index to record the locations of keywords and to judge whether the phrase appears. This index can search for keywords efficiently. In order to rank the results and protect the privacy of relevance score, the relevance score evaluation model is used in searching process on client-side. Also, the special construction of the index makes the scheme dynamic. The data owner can update the cloud data at very little cost. Security analyses and extensive experiments were conducted to demonstrate the safety and efficiency of the proposed scheme.

Cloud computing has become increasingly popular among individuals and enterprises because of the benefits it provides by outsourcing their data to cloud servers. However, the security of the outsourced data has become a major concern. For privacy concerns, searchable encryption, which supports searching over encrypted data, has been proposed and developed rapidly in secure Boolean search and similarity search. However, different users may have different requirements on their queries, which mean different weighted searches. This problem can be solved perfectly in the plaintext domain, but hard to be addressed over encrypted data. In this study, the authors use locality‐sensitive hashing (LSH) and searchable symmetric encryption (SSE) to deal with a privacy preserving weighted similarity search. In the authors’ scheme, data users can generate a search request and set the weight for each attribute according to their requirements. They treat the LSH values as keywords and mix them into the framework of SSE. They use homomorphic encryption to securely address the weight problem and return the top‐k data without revealing any weight information of data users. They formally analysed the security strength of their scheme. Extensive experiments on actual datasets showed that their scheme is extremely effective and efficient.

Internet of Things (IoT) is an increasingly popular technological trend. The operation of IoT needs a strong data-handling capacity, where most of the data are sensor data. Limitations associated with measurement, delays in data updating, and/or the need to preserve the privacy of data can result in the sensor data being uncertain. Thus, one key challenge is “how do we ensure the privacy of data collected from IoT devices, particularly uncertain data, that are being outsourced to the cloud for analysis, storage and archival?”. Searchable encryption scheme is a promising technique that allows the searching over encrypted (uncertain) data stored offshore. In this paper, we propose a secure range search for encrypted data from IoT devices. Specifically, we use homomorphic and order-preserving encryption to encrypt data published by the data owners. We then use the k-dimensional tree to build the data index. Our scheme is designed to ensure the privacy of the dataset, without affecting the efficiency of keyword search on the (encrypted) dataset. We also demonstrate that our scheme can preserve both data and query privacy, as well as evaluating its performance to demonstrate efficiency.

Uncertain data pervades many fields, including environmental monitoring, the monitoring of animal migrations, and urban warfare. Such uncertain data collected by field devices, such as Internet of Things (IoT) and Internet of Battlefield Things (IoBT) devices, may also be encrypted and outsourced to an untrustworthy third party for storage and data sharing such as a cloud server. However, the properties of uncertain data and the complication of operating over encrypted data make the searching schemes more ineffective. In this article, we design an efficient and safe K nearest neighbor (KNN) query scheme for uncertain data stored in semi-trusted cloud servers. We apply the modified homomorphic encryption, which requires two servers to interact and encrypt the uncertain data, and we use the authorized rank method to compute KNN. We protect the security of the data while simultaneously improving the query efficiency. Our detailed security analysis show that our scheme can realize the goal of concealing both the access and the search patterns. Comprehensive experiments are conducted to demonstrate the scheme's performance.

Mobile Crowdsourcing (MC) provides a great platform for people to collect sensing data. Requesters and workers can interact and gain revenue through the platform. For tasks, the requester often has a budget, and the worker also has a price requirement. However, it is a difficult problem to determine the payments which should be accepted by both requesters and workers. In this paper, We propose a complete task assignment mechanism. The first step, we group tasks by Tasks Grouping with Cohesion (TGC), Tasks Grouping with Relevance (TGR) and Tasks Grouping with Similarity (TGS) respectively to form task groups. In the second step, we assign workers to task groups through worker selection mechanisms Workers Selection with Fixed Price (WSFP) and Workers Selection with Changing Price (WSCP). We prove that our mechanism is individual-rational, budget-balance, truthful and computationally efficient. Finally, we evaluate the mechanism through a lot of experiments.

Abstract Asymmetric group key agreement is a cryptographic primitive allowing a group of users to negotiate a common public encryption key while each of them holds a different secret private decryption key. Anyone (including outsiders) with the public encryption key can send encrypted messages to the group members, and then the group members can decrypt the messages. Authenticated key agreement protocols authenticate the identities of users to ensure that only the intended group members can establish a session in which the group members can communicate with each other. Dynamic asymmetric group key agreement concerns about the scenarios such as ad hoc networks in which the group members may join or leave at any given time. In this paper, we propose a one‐round authenticated dynamic protocol for symmetric group key agreement. For efficiency reasons, we employ the identity‐based public‐key cryptography (IB‐PKC) to authenticate users rather than the public key infrastructure and the certificate‐less public‐key cryptography. Our analysis shows that the proposals in the paper can resist active attacks and meet many desirable security attributes. Besides, our protocol allows users to join or leave the group at the same time. Furthermore, our protocol is round‐optimal and has a quite good performance as compared with previous works. Copyright © 2016 John Wiley & Sons, Ltd.

In autonomous environment (such as P2P, ad hoc, social networks and so on), all the rational individuals make independent decisions to maximize their profits. However, many interactions among individuals can be modeled as Prisoner's Dilemma game, which suppresses the emergence of cooperation. In order to provide scalable and robust services in such systems, incentive mechanisms need to be introduced. In this paper, we propose a novel incentive mechanism called recommendation incentive mechanism based on evolutionary dynamics(RIMBED). In our RIMBED system, players who pay an additional cost for recommendation service not only can get the information of the opponents, but also can have a higher probability to interact with cooperative individuals. Using the replicator dynamics equations in evolutionary game theory, we mathematically analyze the robustness and effectiveness of our RIMBED system. Meanwhile, simulation experiments can also validate our mathematical analysis. In our RIMBED system, players have three alternative strategies: always cooperative(ALLC), always defective(ALLD) and rational cooperative(RC). No one strategy can dominate the others forever and all the three strategies can survive in our system. When we bring in population invasion and a small mutation, our system can still work at an excellent level.

In traditional secret image sharing schemes, the participants have the same status, and shadow images are of approximately equal importance, which cannot satisfy some special requirements in real situations. In this paper, we considered the problem of secret image sharing with the weighted threshold access structure, which means different participants can have different status and significance. With this approach, each shadow image has one weight, and the secret image can be reconstructed losslessly if, and only if, the sum of all of the shadow images' weights is no less than the given weight threshold. In our scheme, we constructed the weighted threshold access structure of shadow images using the weighted threshold secret sharing scheme. Then, we used the quantization operation to embed the secret image's information into a host image to generate shadow images with different weights. In the retrieving procedure, a set of shadow images that satisfied the weighted threshold was utilized to restore the distortion-free secret image. Our experimental results confirmed that the proposed scheme was feasible, and both the visual quality of the shadow images and the embedding capacity of the host images were satisfactory. Copyright © 2015 John Wiley & Sons, Ltd.

In a group‐based trust management scheme, peers are partitioned into groups based on chosen characteristics, such as location and interest. The super peer (SP), who is responsible for the storage and distribution of reputation value, has an important role in group‐based trust management. Thus, if the SP is a disguised or malicious peer, serious security problems could occur. To solve these security problems, the authors propose a traceable, group‐oriented, signature scheme with multiple signing policies for trust management. The SP's signature is generated by a designated group called the signature group. In the authors scheme, peers in the signature group will decide whether to generate the signature for the SP based on the SP's reputation, meaning that attackers cannot forge a valid signature. In addition, an outsider also can trace the signers who were involved in generating the signature for reputation valuation.

ABSTRACT Many traditional steganography methods do not disperse and hide secret data smoothly over all the capacity of cover images. Rather, they severely modify part of the cover image(s) to embed secret data, which results in stego images that have poor visual quality. In addition, there are still some secret image‐sharing approaches that cannot reveal the secret image losslessly without pixel expansion or extra storage or restore distortion‐free cover image(s) if they use steganography. In this paper, a novel scheme, which is based on Shamir's (t,n)‐threshold scheme (1979) and Galois Field GF(28) and uses dynamic embedding and least significant bit construction, is proposed to solve the issues mentioned above. Our experimental results showed that the dynamic embedding performance in our scheme was satisfactory and that both the secret image and the cover image can be restored losslessly without pixel expansion or extra storage. Copyright © 2012 John Wiley & Sons, Ltd.