Mobile devices have penetrated our daily life, and so have the fast chargers. However, while improving users' charging efficiency, fast chargers may pose severe security threats. In this work, we present a novel side-channel attack called ChargerWhisper, which exploits the acoustic signals generated by fast chargers to infer users' private information in an on-charging mobile device. In particular, when users perform different activities on the charging device, the fast charger will supply distinct amounts of power outputs to the mobile device. By conducting a deep investigation of the fast charger circuits, we find that the electronic components, i.e., inductors, capacitors, and transformers, vibrate at certain frequencies and generate high-frequency inaudible sounds. Meanwhile, the leaked sound frequencies are highly correlated with electric intensity flowing through these electronic components (charger's output power). To demonstrate the acoustic side-channel attack leveraging fast chargers, we present two specific exploitation scenarios: a website fingerprinting attack to identify users' on-browsing websites, and a PIN inference attack to infers the device's unlock PIN. Extensive evaluations are performed on off-the-shelf devices. The results show that both attacks are robust under diverse usage settings and achieve good performance in inferring users' privacy information. These findings highlight the essential need to address unnoticed privacy leakage risks associated with fast chargers.

Present Advanced Driving Assistance System (ADAS) responds to the dangerous crossing of pedestrians after the occurrence of the incident, occasionally causing severe accidents due to the stringent response window. Inference of pedestrian crossing intention may help vehicles operate in advance and enhance the safety of the vehicle by predicting the crossing probability. Recent studies usually ignore the demand of real-time forecast that required in the realistic driving scenario, and mainly focus on improving the model representation capacity on public datasets by increasing modality and observation time. Consequently, a new framework named EfficientPIE is proposed to predict the pedestrian crossing intention in real time with sole observation of the incident. To achieve reliable predictions, we propose incremental learning based on intention domain to relieve forgetting and promote performance with a progressive perturbation method. Our EfficientPIE outperforms all the SOTA models on two datasets PIE and JAAD, running nearly 7.4x faster than the previously fastest model. Our code is available at https://github.com/heinideyibadiaole/EfficientPIE.

In the field of medical imaging driven by the Internet of Things (IoT), with the rapid growth of the number of medical devices and the widespread application of edge computing (EC) technology, efficient collaborative computing on resource-constrained end medical devices has become the key to improving diagnostic efficiency, thereby bringing a more patient-friendly diagnosis and treatment experience. Computed tomography perfusion (CTP) images play an irreplaceable role in the assessment of brain tissue ischemia in patients with acute ischemic stroke (AIS), with high diagnostic accuracy in identifying ischemic lesions and distinguishing infarction from penumbra, but it has the disadvantages of high radiation dose and high cost. To this end, we propose a CTP image synthesis framework based on edge-end collaboration (EESyn-CTP), which aims to use non-contrast CT (NCCT), CT angiography (CTA), and delayed CTA (CTA+8s) images to synthesize CTP images with arbitrary time to optimize AIS diagnosis. The framework consists of two stages: the pre-training stage on the edge server and the fine-tuning stage on the end device. Specifically, we first deploy a temporal residual generative network, t-UNet, on the edge server for pre-training. This process utilizes multiple CTP images, which share similar perfusion features with CTA, CTA+8s, and NCCT images, to effectively learn the gap in perfusion information between the inputs and outputs. Subsequently, the pre-trained t-UNet model parameters are frozen and broadcast to the edge medical device. A UNet adapter is introduced before the model, and fine-tuning is performed on the adapter weights using real NCCT, CTA, and CTA+8s images as input. This approach facilitates the synthesis of CTP images at arbitrary time points. Finally, experiments on an internal data set showed that the quality of Syn-CTP images synthesized by the EESyn-CTP framework is comparable to that of real CTP images and significantly reduces computation latency and energy overhead.

With the rise of telemedicine and intelligent diagnostics, the efficiency and accuracy of healthcare services have been significantly enhanced. However, the highly sensitive nature of medical images makes protecting patient privacy during transmission and storage a critical challenge. In this paper, we propose a secure medical image encryption scheme based on cross-ring Josephus scrambling and two-dimensional cellular automata, designed to safeguard medical images. First, we introduce a novel two-dimensional chaotic map (2D-CICM) with an expanded parameter range to generate high-quality key sequences for encryption. Next, we design a cross-ring Josephus scrambling algorithm for pixel permutation, where the eliminated pixel is determined by both inter-ring and intra-ring step sizes. Following this, we develop a diffusion mechanism based on interaction rules defined by six types of two-neighbor structures within a cellular automaton framework. To enhance key sensitivity and image-specific security, we also incorporate the 512-bit hash value of the plaintext image to dynamically update the initial keys, ensuring that the encryption key sequences are unique for each image. Comprehensive security analyses and performance evaluations confirm that the proposed scheme provides strong encryption performance and effectively resists common attacks, while maintaining computational efficiency suitable for medical applications.

The local deployment of federated large language models (FLLM) has further advanced the development of edge intelligence. However, the resource constraints of end devices, device heterogeneity, and the non-independent and identically distributed (Non-IID) nature of data pose significant challenges to the application of FLLM. To address this issue, we propose an Adaptive Asynchronous Accelerated FLLM (Tri-AFLLM) algorithm to achieve the efficient utilization of limited resources and improve model accuracy in the edge computing (EC) scenarios. Specifically, Tri-AFLLM first ships an off-the-shelf LLM, i.e., CLIP, to each end device, keeping the backbone parameters frozen and updating only the parameters of the adapter containing two linear transformation layers by using momentum gradient descent (MGD). Next, a toy example is provided to illustrate the necessity of using different numbers of local iterations for heterogeneous devices in resource-constrained environments. Subsequently, the convergence bound of the Tri-AFLLM under a given resource budget is discussed. Then, we formulated the bound into a resource consumption minimization problem with the number of local iterations as the optimization variable under a given model accuracy to mitigate the contribution disparity of local models to the global aggregation. Finally, extensive experiments are conducted to validate the superiority of Tri-AFLLM in terms of resource consumption, model accuracy, and addressing the Non-IID problem.

With the development of Wi-Fi sensing, wireless-based gait recognition has become increasingly important as it supports a wide range of applications (person identification, disease diagnosis, etc.). However, two serious challenges limit the universal deployment of such Wi-Fi vision schemes: i) the limited bandwidth of Wi-Fi severely restricts the granularity of gait recognition, and ii) users non-gait behaviors (e.g., stopping and turning) interfere with the extraction of gait-related features. In this paper, we propose CSID, which can achieve robust gait recognition under the limited bandwidth conditions of commercial Wi-Fi devices. Specifically, we use a neural network to generate super-resolution spectrograms of channel state information (CSI), overcoming the limitation of insufficient Wi-Fi bandwidth. To overcome the challenge of non-gait behavior interference, considering the human-incomprehensible nature of Wi-Fi spectrograms, we adopt cross-domain adversarial training and further extract gait features that are independent of the interference behaviors by learning domain-independent representations. We conducted a large number of experiments in different indoor environments, and the average person identification rate of the CSID system reached 91.6%. These results demonstrate that the CSID system is promising and could be used as a complement to visual person identification systems in the future.

Transferring important files, photos, and other sensitive data between laptops and smartphones has become a routine necessity in daily life. Device pairing acts as the most fundamental need to secure the communication channel between two unconnected devices. Traditional pairing methods leveraging PINs or QR codes require tedious human efforts in the pairing procedures to establish a shared communication key. Nevertheless, these designs are vulnerable to shoulder-surfing attacks in which attackers might record and replay the pairing credentials. It is preferred to have more intuitive pairing designs that minimize users’ overhead in the pairing process while providing secure keys for communication purposes. In this paper, we propose MIPair for laptop-phone pairing by leveraging magnetic induction (MI) signals. MIPair is based on a key observation that changes in the CPU workload of a device cause variations in internal current, thereby inducing changes in surrounding magnetic fields. Moreover, the trends of MI signal variations are highly correlated with CPU workload trends. Thus, users simply need to place a smartphone on the keyboard of a laptop. By randomly altering the workload of the laptop through a stimulation program, the smartphone can capture MI signals with similar changing trends, thereby converting them into similar bit sequences that form the basis of a symmetric key. We propose essential techniques to overcome challenges such as time asynchronization between two devices, unfixed state transition time in MI signal, and shared key distribution from the two similar bit sequences. Our real-world experiments demonstrate reliable pairing as well as robustness against common attacks and high randomness of the generated keys. When generating a 128-bit key, MIPair achieves a successful pairing rate as high as 98% and a usable pairing time of 8.35 seconds.

The escalating threat of hidden GPS tracking devices poses significant risks to personal privacy and security.Featured by their miniaturization and misleading appearances, GPS devices can be easily disguised in their surroundings making their detection extremely challenging.In this paper, we propose a novel side-channel-driven detection system, GPSBuster, leveraging electromagnetic radiation (EMR) emitted by GPS trackers.Our feasibility studies and hardware analysis reveal that unique EMR patterns associated with the tracker's operation, stemming from the quartz oscillator, local oscillator, and mixer in the Mixed-Signal on Chip (MSoC) system.Nevertheless, as a side-channel leakage, EMRs can be extremely weak and suffer from the ambient noise interference, rendering the detection impractical.To address these challenges, we develop the signal processing techniques with noise removals and a dual-dimensional folding mechanism to accumulate the spectrum energy and protrude the EMR patterns with high Signal-to-Noise Ratios (SNR).Our detection prototype, built with a portable HackRF One device, allows users to perform a scan-to-detect manner and achieves an overall success rate of 98.4% on top-10 selling GPS trackers under various testing cases.The maximum detection range is 0.61m.

Internet of Things (IoT) devices play essential roles in both industry and daily scenarios. However, unlike smartphones and computers, IoT devices typically lack conventional user interfaces (UIs) such as keyboards and touchscreens. It renders the traditional user authentication designs, e.g., PINs and patterns, inapplicable. In this article, we proposeTouchAccessthat enables users to unlock an arbitrary IoT device by applying a simple touch. Our design is motivated by the key observation that IoT devices unavoidably generate electromagnetic emanations (EMM) while they are functioning. When the user touches the device, it causes time-varying coupling between those two and generates unique EMMs. Our feasibility studies further reveal that thesehuman-induced EMMsare distinct and strongly correlated with the circuitry properties of the user and the device, but are susceptible to environmental EM noises, thus lowering the authentication accuracy. To address this challenge, we develop signal processing techniques with a Siamese network learning scheme that clears the ambient electromagnetic (EM) noises, extracts robust signal features, and builds noise-resistant classifiers, enabling users to be correctly recognized. A significant advantage ofTouchAccessis that it requires only a low-cost analog-to-digital converter (ADC) to sense the EM signal. We implementTouchAccesson commercial off-the-shelf (COTS) IoT devices, which vary significantly in terms of UIs, sizes, and hardware designs. The performance evaluations show thatTouchAccessachieves an average authentication accuracy as high as 97.85%.

The popularity of IoT devices has penetrated our daily life while posing new challenges in user authentication. Today’s solutions, e.g., passwords, fingerprints, and FaceIDs, primarily rely on specialized sensors or user interfaces to collect user’s identification information, which may not universally exist on heterogeneous IoT devices. In this paper, we propose a novel user authentication method that exploits the Electromagnetic (EM) emanations radiated from IoT devices. Our design is motivated by the observation that human touches on the IoT device can lead to time-varying coupling between these two. Consequently, it impacts the device’s EM emanations that can be picked up by its inertial ADC (analog-to-digital converter) interfaces. We ask the user to tap on the device rhythmically following a self-determined melody, such that the human-coupled EM emanations vary accordingly. We thus extract the rhythm pattern as the user’s secure password named EM-Rhythm . To examine its effectiveness, EM-Rhythm is implemented on a wide range of IoT devices. We show that our scheme achieves authentication accuracy as high as 98.67% with less than three login attempts. Besides, it is robust against various types of attacks and maintains stable performances under various settings. EM-Rhythm also exhibits satisfactory usability in terms of memorability and time consumption.

Graph theory analysis using electroencephalogram (EEG) signals is currently an advanced technique for seizure prediction. Recent deep learning approaches, which fail to fully explore both the characterizations in EEGs themselves and correlations among different electrodes simultaneously, generally neglect the spatial or temporal dependencies in an epileptic brain and, thus, produce suboptimal seizure prediction performance consequently. To tackle this issue, in this article, a patient-specific EEG seizure predictor is proposed by using a novel spatio-temporal-spectral hierarchical graph convolutional network with an active preictal interval learning scheme (STS-HGCN-AL). Specifically, since the epileptic activities in different brain regions may be of different frequencies, the proposed STS-HGCN-AL framework first infers a hierarchical graph to concurrently characterize an epileptic cortex under different rhythms, whose temporal dependencies and spatial couplings are extracted by a spectral-temporal convolutional neural network and a variant self-gating mechanism, respectively. Critical intrarhythm spatiotemporal properties are then captured and integrated jointly and further mapped to the final recognition results by using a hierarchical graph convolutional network. Particularly, since the preictal transition may be diverse from seconds to hours prior to a seizure onset among different patients, our STS-HGCN-AL scheme estimates an optimal preictal interval patient dependently via a semisupervised active learning strategy, which further enhances the robustness of the proposed patient-specific EEG seizure predictor. Competitive experimental results validate the efficacy of the proposed method in extracting critical preictal biomarkers, indicating its promising abilities in automatic seizure prediction.

Dynamic functional connectivity (dFC) network inferred from resting-state fMRI reveals macroscopic dynamic neural activity patterns for brain disease identification. However, dFC methods ignore the causal influence between the brain regions. Furthermore, due to the complex non-Euclidean structure of brain networks, advanced deep neural networks are difficult to be applied for learning high-dimensional representations from brain networks. In this paper, a group constrained Kalman filter (gKF) algorithm is proposed to construct dynamic effective connectivity (dEC), where the gKF provides a more comprehensive understanding of the directional interaction within the dynamic brain networks than the dFC methods. Then, a novel virtual adversarial training convolutional neural network (VAT-CNN) is employed to extract the local features of dEC. The VAT strategy improves the robustness of the model to adversarial perturbations, and therefore avoids the overfitting problem effectively. Finally, we propose the high-order connectivity weight-guided graph attention networks (cwGAT) to aggregate features of dEC. By injecting the weight information of high-order connectivity into the attention mechanism, the cwGAT provides more effective high-level feature representations than the conventional GAT. The high-level features generated from the cwGAT are applied for binary classification and multiclass classification tasks of mild cognitive impairment (MCI). Experimental results indicate that the proposed framework achieves the classification accuracy of 90.9%, 89.8%, and 82.7% for normal control (NC) vs. early MCI (EMCI), EMCI vs. late MCI (LMCI), and NC vs. EMCI vs. LMCI classification respectively, outperforming the state-of-the-art methods significantly.