In this paper, we propose a novel integrated framework for trustworthy artificial intelligence (AI) systems, which deeply fuses privacy computing (PC), blockchain (BC), and large language models (LLMs) into a cohesive unit (TrustFusion AI). LLMs have emerged as significant drivers of digital economic growth and industrial transformation. However, their widespread adoption has raised significant data privacy and security concerns that conventional centralized security approaches have struggled to address. Despite the potential of PC and BC to offer solutions, there is still a lack of an integrated framework that systematically combines these technologies with LLMs to address practical issues of privacy, trust, and compliance. Our proposed TrustFusion AI deep fusion framework views these technologies as a cohesive unit designed to establish an end-to-end trustworthy AI service chain. In this framework, we analyze the synergistic roles of each component: privacy computing secures data confidentiality and invisibility during computation; blockchain serves as a decentralized trust anchor, ensuring the integrity and traceability of data and computational processes; and LLMs serve as the intelligent core, leveraging their robust analytical capabilities while safeguarding privacy. Future research should prioritize investigating the bidirectional enabling mechanism of LLMs and PC/BC. In this paper, we demonstrate the practical application of this framework through two industrial case studies: intelligent manufacturing quality control and enterprise data compliance audits. Furthermore, we address the primary challenges and future research directions concerning technical standardization and the balance between performance and safety within the framework. Our study offers a viable approach for establishing a reliable and sustainable AI data collaboration technology system and presents a novel perspective for future research in this domain.

Artificial Intelligence Generated Content (AIGC), particularly video generation with diffusion models, has been advanced rapidly. Invisible watermarking is a key technology for protecting AI-generated videos and tracing harmful content, and thus plays a crucial role in AI safety. Beyond post-processing watermarks which inevitably degrade video quality, recent studies have proposed distortion-free in-generation watermarking for video diffusion models. However, existing in-generation approaches are non-blind: they require maintaining all the message-key pairs and performing template-based matching during extraction, which incurs prohibitive computational costs at scale. Moreover, when applied to modern video diffusion models with causal 3D Variational Autoencoders (VAEs), their robustness against temporal disturbance becomes extremely weak. To overcome these challenges, we propose SIGMark, a Scalable In-Generation watermarking framework with blind extraction for video diffusion. To achieve blind-extraction, we propose to generate watermarked initial noise using a Global set of Frame-wise PseudoRandom Coding keys (GF-PRC), reducing the cost of storing large-scale information while preserving noise distribution and diversity for distortion-free watermarking. To enhance robustness, we further design a Segment Group-Ordering module (SGO) tailored to causal 3D VAEs, ensuring robust watermark inversion during extraction under temporal disturbance. Comprehensive experiments on modern diffusion models show that SIGMark achieves very high bit-accuracy during extraction under both temporal and spatial disturbances with minimal overhead, demonstrating its scalability and robustness.

Pathological diagnosis is considered the gold standard in cancer diagnosis, playing a crucial role in guiding treatment decisions and prognosis assessment for patients. However, achieving accurate diagnosis of pathology images poses several challenges, including the scarcity of pathologists and the inherent subjective variability in their interpretations. The advancements in whole-slide imaging technology and deep learning methods provide new opportunities for digital pathology, especially in low-resource settings, by enabling effective pathological image classification. In this article, we begin by introducing the datasets, which include both unimodal and multimodal types, as essential resources for advancing pathological image classification. We then provide a comprehensive overview of deep learning-based pathological image classification models, covering task-specific models such as supervised, unsupervised, weakly supervised, and semi-supervised learning methods, as well as unimodal and multimodal foundation models. Next, we review tumor-related indicators that can be predicted from pathological images, focusing on two main categories: indicators that can be recognized by pathologists, such as tumor classification, grading, and region recognition; and those that cannot be recognized by pathologists, including molecular subtype prediction, tumor origin prediction, biomarker prediction, and survival prediction. Finally, we summarize the key challenges in digital pathology and propose potential future directions.

Watermarking technology has gained significant attention due to the increasing importance of intellectual property (IP) rights, particularly with the growing deployment of large language models (LLMs) on billions resource-constrained edge devices. To counter the potential threats of IP theft by malicious users, this paper introduces a robust watermarking scheme without retraining or fine-tuning for transformer models. The scheme generates a unique key for each user and derives a stable watermark value by solving linear constraints constructed from model invariants. Moreover, this technology utilizes noise mechanism to hide watermark locations in multi-user scenarios against collusion attack. This paper evaluates the approach on three popular models (Llama3, Phi3, Gemma), and the experimental results confirm the strong robustness across a range of attack methods (fine-tuning, pruning, quantization, permutation, scaling, reversible matrix and collusion attacks).

The traditional blockchain-based"true-storage"system ignores the provision of"nonexistent proof"when filtering"invalid query requests",and malicious nodes can launch denial of service attacks on designated users.This paper proposes a method for building a verifiable Bloom filter,by which nodes can quickly filter invalid query requests and provide valid evidence proving the data does not exist.In addition,this paper proposes two ways to solve the privacy leakage problem in the proof process:a"hidden verifiable Bloom filter"and"data confusion."The former method ensures that each"nonexistent proof"only leaks one bit of the Bloom filter,reducing the data leakage in the proof process.The latter method reduces the accuracy for users inferring real content from the leaking Bloom filter.The experimental data show that when the proportion of invalid query requests is 35％,the read performance can be improved by approximately 30％;and when this proportion is 95％,the improvement can be more than tenfold.

Dispatching flexible unmanned aerial vehicles (UAVs) to collect data from distributed Internet-of-Things devices (IoTDs) is expected to be a promising technology to support time-critical applications. However, in urban environments, the communication links between the UAV and IoTDs are prone to be frequently blocked by buildings, which severely impairs the freshness of information collected by the UAV. Thus, how to overcome urban building blockages and ensure fresh data collection is quite important but neglected in existing works. In this paper, for keeping the information fresh, we propose to utilize the reconfigurable intelligent surface (RIS) to assist the UAV in mitigating signal propagation impairments caused by building blockages. The formulated optimization problem is minimizing the age of information (AoI) of all IoTDs by jointly optimizing the UAV trajectory, IoTD scheduling and discrete phase shifts of the RIS. It is a mixed-integer non-convex problem as well as lacking the complete channel state information (CSI), thus using conventional optimization methods is intractable. To address this issue, we present an effective and robust deep reinforcement learning (DRL)-based scheme called “SAC-AO-RIS,” where a soft actor-critic (SAC) algorithm with recent-prioritized experience replay is designed for learning highly stable policies of UAV trajectory and IoTD scheduling, and an alternating optimization (AO) algorithm is leveraged for solving RIS phase shifts. Finally, simulation results demonstrate the effectiveness and superiority of our proposed scheme compared with other baseline approaches.

Cloud-edge collaboration, as an emerging computing paradigm, aims to solve the shortcomings of remote transmission of conventional cloud computing. More precisely, it combines the powerful resource service capability of cloud computing with the advantages of low latency and relatively low energy consumption of edge computing to achieve the goal of optimization of various applications. However, with the rapid growth of computation-intensive industrial tasks, the overload problem of edge networks is becoming increasingly serious. Prior studies usually assume that the real-time state of edge resources has been known when selecting the offloading strategy so as to classify and execute tasks, but do not consider the fragmentation and heterogeneity features of edge computing resources. In light of these, we first generalize and model the computing resources of the edge nodes uniformly and then propose new heterogeneous task classification and recognition methods empowered by edge intelligence. We conduct intensive experiments to justify that our proposed design can minimize the data transmission delay caused by repeated computational tasks while saving energy consumption.

The Industrial Internet of Things (IIoT) is deemed a promising direction to drive a new industrial revolution. However, due to the isolation of the existing OT network and IT network, the requirements of low latency, low jitter, and high reliability for transmission and processing of industrial time-sensitive tasks data traffic in IIoT scenarios with strong dynamic and complex topology face a series of non-trivial challenges. In this paper, we propose a hierarchical computing network collaboration architecture for IIoT based on edge/fog computing. Our architecture is built upon the Time-Sensitive Networking (TSN) to flexibly support different requirements of large-scale industrial production applications by constructing the hierarchical computing network collaboration domain, combined with an improved Cyclic Queuing and Forwarding (CQF) scheduling shaper mechanism. We tackle the critical problems of architecture design by presenting three essential components. Moreover, we build and implement our simulation testbed based on Omnet++, and evaluate our design.

As a popular privacy-preserving model training technique, Federated Learning (FL) enables multiple end-devices to collaboratively train Deep Neural Network (DNN) models without exposing local privately-owned data. According to the FL paradigm, resource-constrained end-devices in IoT should perform model training which is computation-intensive, whereas the edge server occupied with powerful computation capability only performs model aggregation. Due to the above unbalanced computation pattern, IoT-oriented FL is time-consuming and inefficient. In order to alleviate the computation burden of end-devices, recent countermeasures introduce the edge server to assist end-devices in model training. However, existing works neither efficiently address the computation heterogeneity across end-devices nor reduce the leakage risk of data privacy. To this end, we propose a Federated Synergy Learning (FedSyL) paradigm which innovatively strikes a balance between training efficiency and data leakage risk. We explore the complicated relationship between the local training latency and multi-dimensional training configurations, and design a uniform training latency prediction method by applying the polynomial quadratic regression analysis. Additionally, we design the optimal model offloading strategy with the consideration of resource limitation and computation heterogeneity of end-devices, so as to accurately assign capability=matched device-side sub-models for heterogeneous end-devices. We implement FedSyL on a real test-bed comprising multiple heterogeneous end-devices. Experimental results demonstrate the superiority of FedSyL on training efficiency and privacy protection.

Surface inspection of industrial equipment defection plays a vital role in real production. Traditional inspection routines require a large number of inspection workers, which not only affects production efficiency but also leads to unreliable results. Computer vision-based detection approaches, e.g., using the deep learning method, have shown great potential in this trend. Specifically, the semantic segmentation algorithm based on Convolutional Neural Network (CNN) can extract relatively complete feature information. And the Transformer, which emerged from the field of Natural Language Processing (NLP), also performs well in maintaining and transmitting semantic information. In light of these, we propose to design a segmentation model called eSwin-UNet, i.e., enhanced Swin-UNet, that leverages the advantages of the CNN and Transformer. It uses multi-scale information fusion to better integrate the feature information in the CNN and Transformer branches. Moreover, it also utilizes deep supervision and makes two branches for collaborative training to further improve accuracy. By testing with the MVTec ITODD dataset, Fl-Score and Jaccard achieve results of 0.7891 and 0.6516 respectively, which outperform most current models.

随着比特币等虚拟数字货币的日益普及与发展,区块链技术受到了研究人员的广泛关注.区块链技术是一种按照时间顺序将区块以链式结构组合而成的分布式数据账本,具有去中心化、可编程、可溯源、不可篡改等特性,在金融领域中的研究尤为广泛.文章面向区块链技术的发展,介绍区块链技术的起源和概述,详细地讨论环签名、零知识证明、数字签名和同态加密等区块链关键技术,综述区块链技术的特点和种类.对区块链技术的应用领域进行概括,重点关注其应用原则和应用领域相关的案例,分析区块链应用当前的发展现状,并对未来区块链技术的发展方向进行分析与预测.

Compared with traditional mobile Ad hoc networks (MANET) and vehicular Ad hoc networks (VANET), unmanned aerial vehicle (UAV) networks have two unique characteristics. First, the UAV network topology changes more frequently and unpredictably. Second, link qualities between UAVs vary significantly with time and space. Existing hop-by-hop routing protocols can neither adapt to the highly dynamic network environment nor fully utilize transmission opportunities in all links with various qualities in UAV networks. In this paper, we propose the Geographic Position based Hopless Opportunistic Routing (GPHLOR) Protocol for UAV networks to solve these problems. GPHLOR combines the advantages of the hopless protocols and the position-based protocols. Benefiting from the idea of hopless, compared with hop-by-hop routing protocols, GPHLOR can fully utilize the transmission opportunities of all links regardless of hop distances and link qualities. While different from existing hopless protocols, GPHLOR selects relays according to each node's geographic position information without caring about the topology, so it can well adapt to the frequent topology change. Moreover, GPHLOR allows each node to calculate its forwarding priority in a distributed manner according to its local position information. It does not rely on frequent network information measurement and exchange, reducing the protocol overhead. Simulation results show that GPHLOR achieves better performance in packet delivery rate, end-to-end delay, throughput, and overhead compared with existing routing protocols in UAV networks.

Blockchain is multi-centralized, immutable and traceable, thus is very suitable for distributed storage, privacy and security management in IoTs. However, most existing researches focus on the integration of public blockchain and IoTs. In fact, problems such as slow consensus, low transmission throughput, and completely open storage on the public blockchain are intolerable in IoT scenarios. Although consortium blockchain represented by Hyperledger Fabric has improved the transmission rate, its data security completely relies on the PKI-based certificate mechanism, resulting in transmission inefficiency and privacy leakage. In this paper, a key-derived Controllable Lightweight Secure Certificateless Signature (CLS2) algorithm is proposed to significantly improve the transmission efficiency and keep similar computation overhead of consortium blockchain. Compared with the existing certificateless signatures, CLS2 achieves more secure transactions, whose controllable anonymity and key-derived mechanism not only prevents public key replacement attacks and forged signature attacks, but also supports hierarchical privacy protection. Armed with CLS2, we design a consortium blockchain security architecture based on Hyperledger Fabric and edge computing. To the best of our knowledge, this is the first implementation of certificateless signature in consortium blockchain. We formally prove the security of our schemes in the random oracle model. Specifically, the security of the proposed scheme is reduced to the Elliptic curve discrete logarithm problem (ECDLP). Security analysis and experiments in IoT scenarios verify the feasibility and effectiveness of CLS2.

深度神经网络(deep neural network,DNN)已经广泛应用于各种智能应用,如图像和视频识别.然而,由于DNN任务计算量大,资源受限的物联网(Internet of things,IoT)设备难以本地单独执行DNN推断任务.现有云协助方法容易受到通信延迟无法预测和远程服务器性能不稳定等因素的影响.一种非常有前景的方法是利用IoT设备协作实现分布式、可扩展DNN任务推断.然而,现有工作仅研究IoT设备同构情况下的静态拆分策略.因此,迫切需要研究如何在能力异构且资源受限的IoT设备间自适应地拆分DNN任务,协作执行任务推断.上述研究问题面临2个重要挑战:1)DNN任务多层推断延迟难以准确预测;2)难以在异构动态的多设备环境中实时智能调整协作推断策略.为此,首先提出细粒度可解释的多层延迟预测模型.进一步,利用进化增强学习(evolutionary reinforcement learning,ERL)自适应确定DNN推断任务的近似最优拆分策略.实验结果表明:该方法能够在异构动态环境中实现显著DNN推断加速.

Quantitative research of digital currency is important in the financial sector. Existing quantitative research mainly focuses on pairs trade, multifactor models, and investment portfolios. Investment portfolios refer to the allocation of funds to different types of financial products to minimise investment risk when expected returns can be obtained, or maximise returns on investment when investment risks are controllable. Herein, we propose a ranking-based digital currency investment strategy framework for investment portfolios in the digital currency market. The framework mainly involves selecting digital currency attributes, pre-processing historical data, exporting the ranking model of the investment portfolio strategy, and parameter optimisation.

As a dominant edge intelligence technique, Federated Learning (FL) can reduce the data transmission volume, shorten the communication latency and improve the collaboration efficiency among end-devices and edge servers. Existing works on FL-based edge computing only take device- and resource-heterogeneity into consideration under a fixed loss-minimization objective. As heterogeneous end-devices are usually assigned with various tasks with different target accuracies, task heterogeneity is also a significant issue and has not yet been investigated. To this end, we propose a Customized FL (CuFL) algorithm with an adaptive learning rate to tailor for heterogeneous accuracy requirements and to accelerate the local training process. We also present a fair global aggregation strategy for the edge server to minimize the variance of accuracy gaps among heterogeneous end-devices. We rigorously analyze the convergence property of the CuFL algorithm in theory. We also verify the feasibility and effectiveness of the CuFL algorithm in the vehicle classification task. Evaluation results demonstrate that our algorithm performs better in terms of the accuracy rate, training time, and fairness during aggregation than existing efforts.

Change-Encryption is a distributed layer-by-layer symmetric encryption technology combined with blockchain technology and cryptography. The approach relies on the identity information of all the nodes in the blockchain, and the time, space, and other attributes are used as encryption parameters to enable repeated encryption. In this work, the identity information of the data sender and receiver is considered as the key, and the spatial and identity information values of the node are considered as dynamic encryption parameters to increase the uncertainty of decryption, thereby solving the data security problem of the blockchain technology during communication. In addition, the self-identification principle of the data receiver, reduction principle of the data sender, and two unique countermeasures employed in the Change-Encryption are introduced in detail. The findings demonstrate the value of the proposed approach.

In distributed Internet of Things (IoTs), channel hopping (CH) is an effective scheme for neighbor nodes to achieve blind rendezvous over common available channels and to establish communication links. When nodes are unaware of each other’s local clocks and the global channels and have no pre-assigned CH strategies or identifiers (IDs), it is particularly challenging to guarantee blind rendezvous within a finite period of time, which has not been solved yet by using only one radio. In this paper, we propose a novel quaternary-encoding-based CH (QECH) algorithm to tackle the above issue. The QECH algorithm encodes a randomly selected channel into a quaternary string according to the 6B/8B encoding. We also append a common prefix string as well as the randomly selected channel before the quaternary string to guarantee overlaps in the asynchronous scenario. For all kinds of quaternary digits, we construct four mutually co-prime numbers to enumerate all possible combinations of the common available channels. We theoretically analyze the deterministic rendezvous principle and the upper bounded rendezvous latency of the QECH algorithm. We also verify the effectiveness of the QECH algorithm through extensive simulations. Evaluation results show the superiority of the QECH algorithm in terms of rendezvous latency.

Delay-Tolerant Networks (DTNs) are wireless networks that often experience temporary, even long-duration partitioning. Current DTN researches mainly focus on pure delay-tolerant networks that are extreme environments within a limited application scope. It motivates the identification of a more reasonable and valuable DTN architecture, which can be applied in a wider range of environments to achieve interoperability between some networks suffering from frequent network partitioning, and other networks provided with stable and high speed Internet access. Such hybrid delay-tolerant networks have a lot of applications in real world. A novel and practical Cache-Assign-Forward (CAF) architecture is proposed as an appropriate approach to tie together such hybrid networks to achieve an efficient and flexible data communication. Based on CAF, we enhance the existing DTN routing protocols and apply them to complex hybrid delay-tolerant networks. Simulations show that CAF can improve DTN routing performance significantly in hybrid DTN environments.

Mobile IP is one of the dominating protocols that enable a mobile node to remain reachable while moving around in the Internet. However, it suffers from long handoff latency and route inefficiency. In this article, we present a novel distributed mobility management architecture, ADA (Asymmetric Double-Agents), which introduces double mobility agents to serve one end-to-end communication. One mobility agent is located close to the MN and the other close to the CN. ADA can achieve both low handoff latency and low transmission latency, which is crucial for improvement of user perceived QoS. It also provides an easy-to-use mechanism for MNs to manage and control each traffic session with a different policy and provide specific QoS support. We apply ADA to MIPv6 communications and present a detailed protocol design. Subsequently, we propose an analytical framework for systematic and thorough performance evaluation of mobile IP-based mobility management protocols. Equipped with this model, we analyze the handoff latency, single interaction delay and total time cost under the bidirectional tunneling mode and the route optimization mode for MIPv6, HMIPv6, CNLP, and ADA. Through both quantitative analysis and NS2-based simulations, we show that ADA significantly outperforms the existing mobility management protocols.