Transfer-based adversarial attacks are key for evaluating the robustness of deep neural networks (DNNs) in black-box settings, yet their effectiveness is often constrained by limited cross-model transferability. Existing feature-level approaches typically rely on single-layer attention guidance or static perturbation patterns, which restrict adaptability across diverse architectures. In this work, we introduce a unified adversarial framework, named Multi-layer Attention-guided Spatial Transformations (MAT), to exploit class-discriminative cues from multiple feature layers to craft highly transferable adversarial examples. MAT integrates Multi-layer Attention Fusion (MAF) to capture complementary low-level and high-level semantics from multiple intermediate layers, Attention-guided Augmentation (AGA) to selectively perturb non-critical regions while preserving semantic integrity, and Spatial Random Transformation (SRT) to introduce stochastic spatial augmentations to diversify patterns during optimization. Unlike prior methods that use static or layer-specific attention, MAT dynamically adapts feature guidance to the architecture and task, which enhances generalization. We evaluate MAT against eleven state-of-the-art (SOTA) transfer-based attacks across nine CNN-based and Transformer-based architectures on ImageNet. Comprehensive experiments demonstrate that MAT consistently outperforms eleven state-of-the-art transfer-based attacks in both white-box and black-box settings, including against adversarially trained and input preprocessing-based defensive models, while maintaining higher semantic similarity to the original inputs. It highlights the superior adversarial robustness and excellent adaptability of MAT in adversarial machine learning. Our code is available athttps://github.com/dislab-gzhu/MAT.

With the rapid development of sixth-generation (6G) communication technology, global communication networks are moving towards the goal of comprehensive and seamless coverage. In particular, low earth orbit (LEO) satellites have become a critical component of satellite communication networks. The emergence of LEO satellites has brought about new computational resources known as theLEO satellite edge, enabling ground users (GU) to offload computing tasks to the resource-rich LEO satellite edge. However, existing LEO satellite computational offloading solutions primarily focus on optimizing system performance, neglecting the potential issue of malicious satellite attacks during task offloading. In this paper, we propose the deployment of LEO satellite edge in an integrated satellite-terrestrial networks (ISTN) structure to supportsecurity-sensitive computing task offloading. We model the task allocation and offloading order problem as a joint optimization problem to minimize task offloading delay, energy consumption, and the number of attacks while satisfying reliability constraints. To achieve this objective, we model the task offloading process as a Markov decision process (MDP) and propose a security-sensitive task offloading strategy optimization algorithm based on proximal policy optimization (PPO). Experimental results demonstrate that our algorithm significantly outperforms other benchmark methods in terms of performance.

We study online data analytics with differential privacy (DP) in decentralized settings. Specifically, online data analytics with local DP protection is widely adopted in real-world applications. Despite numerous endeavors in this field, significant gaps in utility and functionality remain when compared to its offline counterpart. We present an optimal, streamable mechanism:ExSub, for local DP sparse vector estimation. The mechanism enables a range of online analytics on streaming binary vectors, including multi-dimensional binary, categorical, or set-valued data. By leveraging the negative correlation of occurrence events in the sparse vector, we attain an optimal error rate under local privacy constraints, only requiring streamable computations. To surpass the error barrier of local privacy, we also studyExSubrandomizer in the newly emerging (single-message) shuffle model of DP, and provide nearly-tight privacy amplification bounds therein. Additionally, we leverage the online shuffle model that independently permutes users' messages at each timestamp, to design a simplified randomization strategy that can approximately reach Gaussian accuracy in central DP. Through experiments with both synthetic and real-world datasets,ExSubmechanism in the local model have been shown to reduce error by$40\%-60\%$compared to SOTA approaches. TheExSubin the shuffle model can further reduce over$85\%$error, and the online shuffle protocol reduces over$99.7\%$error.

In modern information systems, real-time streaming data are generated in various vertical application scenarios, such as industrial security cameras, household intelligent devices, mobile robots, and among others. However, these low-end devices can hardly provide real-time and accurate data analysis functionalities due to their limited onboard performances. Traditional centralized server computing also suffers from its prolonged transmission latency, resulting in huge response time. To deal with this problem, this article presents a novel distributed computation intelligent system with nearby edge devices, abbreviated as EdgeStreaming, to facilitate rapid and accurate analysis of streaming data. First, we thoroughly explore the available edge devices surrounding the terminal to generate an internally interconnected edge network. This edge network real-time perceives and updates the internal resource status of each edge device, such as computational and storage resources. Dynamic allocation of external computational or storage demands can be made based on the current load of individual edge devices. Consequently, the streaming data perceived by external terminal devices can be transmitted in real-time to any edge gateway. The edge network employs a well-designed task scheduling strategy to partition and allocate streaming data processing demands to one or multiple edge devices. Additionally, it customizes computational requirements judiciously, for instance, by utilizing model compression to expedite computation speed. We deployed an edge network comprising multiple Raspberry Pis, NVIDIA Jetson Nano, and Jetson NVIDIA TX2 devices, successfully achieving real-time analysis and detection of video streaming data. We believe our work provides new technological support for the real-time processing of streaming data.

3D face reconstruction from multiple in-the-wild images in an unsupervised manner poses a significant challenge, primarily due to the pervasive presence of Intrinsic and Extrinsic inconsistencies in facial features. To tackle this, we introduce a novel set of IE-aware consistency losses designed to effectively mitigate these inconsistencies. Our Local Alignment Loss employs neighborhood search techniques to identify and optimize consistent pixel information, thereby reducing intrinsic inconsistencies. In parallel, our Region Subset Selection Loss filters out regions where significant discrepancies exist between the input and reconstructed images, effectively alleviating extrinsic inconsistencies. Extensive experimental results validate the effectiveness of our IE-aware consistency losses in reconstructing detailed 3D facial geometry from images captured in uncontrolled environments.

Online data analytics with local privacy protection is widely adopted in real-world applications. Despite numerous endeavors in this field, significant gaps in utility and functionality remain when compared to its offline counterpart. This work demonstrates that private data analytics can be conducted online without excess utility loss, even at a constant factor. We present an optimal, streamable mechanism for local differentially private sparse vector estimation. The mechanism enables a range of online analytics on streaming binary vectors, including multi-dimensional binary, categorical, or set-valued data. By leveraging the negative correlation of occurrence events in the sparse vector, we attain an optimal error rate under local privacy constraints, only requiring streamable computations during the input’s data-dependent phase. Through experiments with both synthetic and real-world datasets, our proposals have been shown to reduce error rates by 40% to 60% compared to SOTA approaches.

Satellite communication networks have attracted widespread attention for seamless network coverage and collaborative computing. In satellite-terrestrial networks, ground users can offload computing tasks to visible satellites that with strong computational capabilities. Existing solutions on satellite-assisted task computing generally focused on system performance optimization such as task completion time and energy consumption. However, due to the high-speed mobility pattern and unreliable communication channels, existing methods still suffer from serious privacy leakages. In this paper, we present an integrated satellite-terrestrial network to enable satellite-assisted task offloading under dynamic mobility nature. We also propose a privacy-preserving task offloading scheme to bridge the gap between offloading performance and privacy leakage. In particular, we balance two offloading privacy, called the usage pattern privacy and the location privacy, with different offloading targets (e.g., completion time, energy consumption, and communication reliability). Finally, we formulate it into a joint optimization problem, and introduce a deep reinforcement learning-based privacy-preserving algorithm for an optimal offloading policy. Experimental results show that our proposed algorithm outperforms other benchmark algorithms in terms of completion time, energy consumption, privacy-preserving level, and communication reliability. We hope this work could provide improved solutions for privacy-persevering task offloading in satellite-assisted edge computing.

In many mobile applications, user-generated data are presented as set-valued data. To tackle potential privacy threats in analyzing these valuable data, local differential privacy has been attracting substantial attention. However, existing approaches only provide sub-optimal utility and are expensive in computation and communication for set-valued data distribution estimation and heavy-hitter identification. In this paper, we propose a utility-optimal and efficient set-valued data publication method (i.e.,Wheel mechanism). On the user side, the computational complexity is only$O(\min \lbrace m\log m, m e^\epsilon \rbrace )$and communication costs are$O(\epsilon +\log m)$bits, where$m$is the number of items,$d$is the domain size and$\epsilon$is the privacy budget, while existing approaches usually depend on$O(d)$or$O(\log d)$($d \gg m$). Our theoretical analyses reveal the estimation errors have been reduced from the previously known$O(\frac{m^{2} d}{n\epsilon ^{2}})$to the optimal rate$O(\frac{m d}{n\epsilon ^{2}})$. Additionally, for heavy-hitter identification, we present a variant of the Wheel mechanism as an efficient frequency oracle, entailing only$O(\sqrt{n})$computational complexity. This heavy-hitter protocol achieves an identification bar of$\tilde{O}(\frac{1}{\epsilon }\sqrt{\frac{m}{n} \log d})$, reducing by a factor of$\sqrt{m}$relative to existing protocols. Extensive experiments demonstrate our methods are 3-100x faster than existing approaches and have optimized statistical efficiency.

With the rapid development of scientific and technological revolution, the Internet of Thing systems have been widely deployed in many vertical domains such as smart cities, smart grids, smart agriculture, smart medical healthcare, etc. In these scenarios, it is an urgent problem to provide a high-effect IoT solution to support high computation ability with strong data security and privacy constraints. In this paper, we propose a novel secure data computing platform based on blockchain crowdsourcing computation. Each data sample is transmitted to its private blockchain for secure data storage. To support a high computation ability, we present a powerful edge network constituted of nearby edge gateways, edge devices, etc. Thus, these data on the blockchain can be partitioned and allocated to these edge nodes in a crowdsourcing way. We also deploy different AI tools to meet the requirements of vertical applications. Finally, we evaluate our system performance with a real crowdsourced blockchain computation platform to show our effectiveness in vertical application, e.g., face detection.

Deep neural networks (DNNs) are increasingly used as the critical component of applications, bringing high computational costs. Many practitioners host their models on third-party platforms. This practice exposes DNNs to risks: A third party hosting the model may use a malicious deep learning framework to implement a backdoor attack. Our goal is to develop the realistic potential for backdoor attacks in third-party hosting platforms. We introduce a threatening and realistically implementable backdoor attack that is highly stealthy and flexible. We inject trojans by hijacking the built-in functions of the deep learning framework. Existing backdoor attacks rely on poisoning; its trigger is a special pattern superimposed on the input. Unlike existing backdoor attacks, the proposed sequential trigger is a specific sequence of clean image sets. Moreover, our attack is model agnostic and does not require retraining the model or modifying the parameters. Its stealthy is that injecting trojans will not change the model’s prediction for a clean image, so existing backdoor defenses cannot detect it. Its flexibility lies in that adversary can remodify the trojan behavior at any time. Extensive experiments on multiple benchmarks with different frameworks demonstrate that our attack achieves a perfect success rate (up to 100%) with minimal damage to model performance. And we can inject multiple trojans which do not affect each other at the same time, trojans hidden in the framework make a universal backdoor attack possible. Analysis and experiments further show that state-of-the-art defenses are ineffective against our attacks. Our work suggests that backdoor attacks in the supply chain need to be urgently explored.

Bridging the advantages of differential privacy in both centralized model (i.e., high accuracy) and local model (i.e., minimum trust), the shuffle privacy model has potential applications in many privacy-sensitive scenarios, such as mobile user data aggregation and federated learning. Since messages from users are anonymized by semi-trusted shufflers (e.g., anonymous channels, edge servers), every user could hide message among other users’ messages and inject only part of noises (a.k.a. privacy amplification). However, existing works assume that the participating user population is known in advance, which is unrealistic for dynamic environments (e.g., mobile computing, vehicular networks). In this work, we study the shuffle privacy model with a random participating population, and give privacy amplification bounds for population size with commonly encountered binomial, Poisson, sub-Gaussian distribution and etc. For further improving accuracy, we formulate and derive optimal dummy sizes for both non-adaptive and adaptive dummies. Finally, to break the error barrier due to the constraint of sending one single message per user, we design a multi-message shuffle private protocol supporting random population. Experiment results show that our approaches reduce more than 60% error when compared to the local model and naive approaches. We hope this work provides tailored solutions of shuffle privacy for dynamic mobile/distributed computing.

A surface slicing planning method based on the K-means clustering algorithm and improved fruit fly optimization algorithm (FOA) is proposed to address the efficiency problem in the surface processing of special shaped workpieces. The NURBS surface reconstruction is performed on the complex surface of the selected workpiece, and the K-means clustering algorithm with the curvature-distance factor is used to subdivide the surface, and the FOA with the hopping strategy is used to plan the optimal connection path for each subdivision with different starting points. This improves the local search capability of the FOA by improving the mixing of different classes that occurs when the surface is subdivided. Finally, the proposed method is demonstrated to be effective for surface slicing by simulating the slicing plan on the constructed heterogeneous workpiece surface, and the shortest connectivity paths are obtained for different starting points.

Indoor-outdoor (IO) detection provides very useful hints for a mobile device to perform context-aware services. To that end, GPS presents a viable solution by relating a device's IO status with its positioning performance, which depends on the device's exposure to the open sky. This approach, however, is prohibitively expensive in terms of energy consumption and response time. Recent work has thus been focused on exploiting low-energy sensors such as light, cellular, and magnetic sensors to infer the IO status indirectly, at the cost of reduced adaptability or manual labeling effort. In this article, we propose a new method to address these problems. Our method, called SatProbe, reverts to the GPS approach for its directness and robustness, but avoids its drawback by extracting only the number of visible satellites from the raw GPS data, instead of going through extensive computation to obtain a final position. This metric provides a clear indicator of the IO status, yet can be obtained with great efficiency. Experiments on 79 raw GPS traces with 2595 detection points across a variety of environments show that SatProbe produces a 14.2 percent improvement in detection accuracy, with a 98.8 percent reduction in both energy use and detection time, in comparison with the standard GPS method.

The past few years have witnessed a rapid growth of stationless bike sharing services. The service allows the bikes to be dropped off freely and to be found through GPS localization. In practice, the bikes are often parked in close proximity to buildings, where GPS accuracy suffers, making bike search a challenging task. This article proposes a novel approach to addressing this problem. Inspired by multi-antenna systems, our method tries to collect GPS signals from multiple distributed bikes, by organizing a group of bikes into a network, called a BikeGPS network. Formed by pedestrian users who opportunistically measure inter-bike distance via radio sensing and step tracking, the generated network permits one to map all the nodes’ satellite range measurements into a single lead node ’s view. By considering both signal and geometry properties of satellite raw measurements, and using an asynchronous coarse time navigation algorithm, the lead node can accurately derive the locations of all the network nodes. Experiments in real-world scenarios show that BikeGPS significantly improves the localization performance, in terms of both accuracy and solution availability, compared with the naive GPS approach and a high-level cooperative localization method.

The past few years have seen a surge of stationless bike sharing services in many modern cities. The service allows the bikes to be dropped off freely, and to be found through GPS localization. For maximum convenience, the bikes are often parked in close proximity to the buildings, where GPS may perform poorly, making bike search a challenging task. This paper proposes a novel approach to addressing this problem. Inspired by multi-antenna systems, our method tries to collect GPS signals from multiple distributed bikes, by organizing a group of bikes into a network, called the BikeGPS. Formed by pedestrian users who opportunistically measure interbike distance via radio sensing and step tracking, the generated network permits one to map all the nodes' satellite range measurements into a single lead node's view. By considering both signal and geometry properties of satellite raw measurements, and using an asynchronous coarse time navigation algorithm, the lead node can accurately derive the locations of all the network nodes. Real-world experiments show that BikeGPS significantly improves the localization performance, in terms of both accuracy and solution availability, compared with the naive GPS approach and a high-level cooperative localization method.

Indoor-outdoor (IO) detection provides very useful hints for a mobile device to perform context-aware services. To that end, GPS presents a viable solution by relating a device's IO status with its positioning performance, which depends on the device's exposure to the open sky. This approach, however, is prohibitively expensive in terms of energy consumption and response time. Recent work has thus been focused on exploiting low-energy sensors such as light, cellular, and magnetic sensors to infer the IO status indirectly, at the cost of reduced adaptability or explicit user involvement. In this paper, we propose an improving solution to this problem. Our method, called SatProbe, reverts to the GPS approach for its directness and robustness, but avoids its drawback by extracting only the number of visible satellites from the raw GPS data, instead of going through extensive computation to obtain a final position. This metric provides a clear indicator of the IO status, yet can be obtained with great efficiency. Experiments on 79 raw GPS traces with 2595 detection points across a variety of environments show that SatProbe produces higher detection accuracy than previous solutions, with more than an order of magnitude reductions in energy consumption and detection time.

Stationless bike sharing systems allow the bikes to be dropped off freely and to be found through GPS localization. Such flexibility has made it highly popular in an increasing number of cities. However, GPS often performs poorly in urban areas with dense high-rise building, making bikes search a challenging task. This paper proposes a novel cooperative GPS approach to address the problem. The method first organizes a group of shared bikes into a network, called a BikeNet, in a crowdsourcing manner. Then the constructed network maps all the nodes' GPS raw measurements to a lead node's view, to determine an accurate GPS location for each node. Experiment results show that BikeNet improves the localization accuracy by 2.96x and 4.8x, compared with the classic GPS method.

This article presents a Light Intensity--based Positioning System (LIPS) for indoor environments. The system uses off-the-shelf light-emitting diode lamps as signal sources and light sensors as signal receivers. The design is inspired by the observation that a light sensor has deterministic sensitivity to both the distance and incident angle of a light signal, an under-utilized feature of photodiodes now widely found on mobile devices. We develop a stable and accurate light intensity model to capture the phenomenon, based on which a new positioning principle, Multi-Face Light Positioning , is established that uses three collocated sensors to uniquely determine the receiver’s position, assuming merely a single source of light. We have implemented a prototype on both dedicated embedded systems and smartphones. Experimental results show average positioning accuracy within 0.4m across different environments, with high stability against interferences from obstacles, ambient lights, temperature variation, and so on.

Equipped with accelerometers and gyroscopes, modern smartphones provide an appealing approach to infrastructure-free navigation for vehicles in indoor environments (for example parking garages). However, a smartphone-based inertial navigation system (INS) faces two serious problems. First, it is subject to errors that accumulate over time rather quickly, which may grow to a level that renders the navigation meaningless. Second, without human input or external references, the smartphone can hardly infer its initial position/velocity, which is the basis for distance calculation, since all that a smartphone can learn is its acceleration. This raises a practical concern, as users often need to start indoor navigation precisely when they are uncertain of their current whereabouts. In this paper, we present Bumping , a Bump-Aided Inertial Navigation method that significantly alleviates the above two problems. At the core of this method is a Bump Matching algorithm, which exploits the position information of the readily available speed bumps to provide useful references for the INS. The proposed method is easy to implement, requires no infrastructures, and incurs nearly zero extra energy. We conducted real experiments in tree parking garages of different environmental characteristics. The Bumping method produces an average position error of 4-5 m in these scenarios, improving the accuracy by up to 87.1 percent, compared to the basic inertial navigation method.