The goal of this work is to adapt Segment Anything Models (SAM) into crack segmentation tasks via automatic label generation, thus eliminating manual annotation cost. In this regard, an intuitive approach is to extract edges of crack samples and generate labels via the dilation and erosion processes for fine-tuning SAM. However, this simple solution cannot guarantee the quality of generated labels, as crack regions will be corrupted due to the imperfect edge detection. To this end, this paper proposes CoGenSAM, a novel Codebook-interactive Generative Labeling framework that enables an annotation-free SAM fine-tuning. To achieve this, in the first stage, we pre-train a vector-quantized variational auto-encoder (VQVAE) by reconstructing the synthesized crack-like structures for learning crack-aware priors within the codebook. In the second stage, these priors help another VQVAE serve as the restoration model to restore the randomly corrupted structures into uncorrupted ones. Specifically, we propose the crack-aware contrastive-interaction to maximize the mutual information with the above priors via codebook interaction. Then, high-quality labels can be generated by restoring corrupted labels from edge detection, contributing to an annotation-free SAM fine-tuning. We collect a new dataset, Bridge2025, to address the limited availability of related bridge-oriented benchmarks. Experiments show that our performance is close to fully-supervised methods.

Large language models (LLMs) are increasingly deployed in edge computing environments to reduce latency and preserve privacy. However, their inference process presents fundamental challenges for resource-constrained IoT devices. LLM inference involves computationally asymmetric stages: parallelizable prompt processing and sequential token decoding. This asymmetry creates deployment bottlenecks where IoT devices lack capacity for prompt processing while edge nodes suffer from inefficient sequential decoding. This paper presentsE$^{2}$LLM, an efficient distributed inference framework for large language models in heterogeneous edge-IoT environments.E$^{2}$LLMleverages high-capacity edge devices for structural planning and introduces auxiliary lightweight models to generate segment-specific key-value (KV) caches. These minimal inference artifacts enable collaborative parallel decoding across IoT devices without requiring full model instantiation. The framework employs static-dynamic KV cache separation to minimize communication overhead while maintaining semantic coherence through structure-guided coordination. Extensive evaluation on realistic edge testbeds demonstrates significant performance improvements. Under diverse deployment settings,E$^{2}$LLMachieves 74%–87.7% end-to-end latency reduction compared with several state-of-the-art baselines, while maintaining comparable generation quality; meanwhile, it also delivers a 34.6%–72.2% reduction in communication overhead, improves 9-12 × in energy efficiency. The framework exhibits strong scalability under bandwidth-limited conditions, enabling efficient LLM deployment across heterogeneous edge-IoT environments.

Deep neural networks (DNNs) have substantially achieved high predictive accuracy in many vision tasks. However, we find that they are poorly calibrated for crack recognition tasks, as these DNNs tend to produce both under-confident and over-confident predictions in such safety-critical applications, thereby limiting their practical use in real-world scenarios. To address this issue, we propose a novel attack-inspired calibration loss (AICL) that explicitly regularizes class probabilities to be better confidence estimation. Specifically, we first propose the attack-inspired correctness estimation method (ACE) that aims to estimate the correctness degree of each sample via adversarial attacks. Then, we propose Correctness-aware Distribution Guidance, which starts from a distribution perspective that enforces the ordinal ranking of the predicted confidence referring to the estimated correctness degree. The proposed method can be conveniently implemented on top of any DNNs-based crack recognition model by serving as a plug-and-play loss function. To address the limited availability of related benchmarks, we collect a fully annotated dataset, namely, Bridge2024, which involves inconsistent cracks and noisy backgrounds in real-world bridges. Our AICL outperforms the state-of-art calibration methods on various benchmark datasets including CRACK2019, SDNET2018, and our BRIDGE2024.

Split learning (SL) is widely regarded as a promising distributed machine learning framework with superior privacy-preserving properties, lower communication and computation costs. However, in real Internet of Things (IoT) scenarios, existing SL may not perform well because the local data of IoT devices often do not follow the same distribution. This leads to the model continuously adapting to the current data distribution in each training epoch, resulting in a catastrophic forgetting phenomenon. Existing methods typically attempt to add raw or generated data from previous devices in the current training epoch to review knowledge, but direct access to the local data of other devices carries serious privacy risks. Data augmentation techniques based on generative networks often have poor robustness and increase the computation cost on the device side. To address these challenges, we propose a new SL framework called SL without Forgetting (SLwF). To mitigate catastrophic forgetting without accessing any previous data, we propose a contrastive learning-based training method that leverages current training data to review previous knowledge, and learn new knowledge better. Furthermore, we adopt an exponential moving average (EMA)-based model update strategy to preserve lost knowledge, further alleviating the forgetting problem. We implement the SLwF framework in real IoT scenarios and extensively evaluated its performance using four publicly available datasets. Compared to other related research (e.g., IoTSL), SLwF performs better in terms of final accuracy and robustness while avoiding excessive device energy consumption.

Collaborative sensing has emerged as a novel sensing paradigm, entailing multi-sensor data sharing and multimodal modeling to collaboratively understand sensing behaviors. However, current solutions, i.e., data-level and decision-level fusion methods, fall short of generality, expert knowledge, and holistic/chronic perspective. In this paper, we proposeLLMCoSento revisit collaborative sensing with Large Language Models (LLMs). Specifically,LLM-CoSendesigns a semantic-level fusion approach for inference results for collaborative sensing. Such an approach is characterized by its generality, making it applicable to any heterogeneous devices, and its expert knowledge incorporation, which provides chronic, holistic, and insightful perspectives on the inference results. Regarding inference absence challenges, we propose a personalized model design method to constrain inference time, and a voting-based two-pass prompt engineering strategy for token completion. Regarding inference error challenges, we propose an accuracy restoration strategy for personalized models, and a two-level error estimator coupled with self-correction. Experimental results of human digital system use case on four corresponding benchmark datasets showLLM-CoSencan decrease inference absence by 72.83% and inference errors by 7.65% on average.

Proper and effective toothbrushing technique is crucial for maintaining oral health. However, there are often limited opportunities for individuals to receive specific training in toothbrushing posture in their daily lives. In this article, we propose Hygiea+, a convenient, energy-efficient, and highly accurate toothbrushing monitoring system based on wrist-worn wearables. By leveraging inertial measurement units (IMUs) in wrist-worn devices for gesture sensing, Hygiea+ enables users to accurately and efficiently monitor their toothbrushing activities without any modifications to the toothbrush. We propose a number of novel techniques to achieve the goal of high sensing accuracy and energy efficiency. To reduce the energy consumption of continuous IMU sampling, we model the sensing problem as a Markov process and design a partially observable Markov decision process (POMDP)-based adaptive sampling strategy to dynamically adjust the sampling frequency. To achieve high sensing accuracy, we first propose a novel signal preprocessing method to mitigate variations resulting from different toothbrush types and user habits. Then, we propose a deep reinforcement learning-based data distillation mechanism to extract key segments from continuous toothbrushing actions, thus reducing the impact of redundant data and noise. In the classification stage, we design an attention-based long short-term memory (AT-LSTM) network for fine-grained toothbrushing posture recognition. In addition, to address the accuracy degradation of new users, we adopt the common but effective fine-tuning method to alleviate the data collection burden on new users. Finally, we connect advanced large language models (LLMs) to provide users with necessary feedback on toothbrushing behavior and health recommendations. Extensive experiments using both manual and electric toothbrushes demonstrate Hygiea+ achieves up to 98.8% accuracy in toothbrushing posture recognition while maintaining superior energy efficiency.

Recently proposed split learning (SL) is a promising distributed machine learning paradigm that enables machine learning without accessing the raw data of the clients. SL can be viewed as one specific type of serial federation learning. However, deploying SL on resource-constrained Internet of Things (IoT) devices still has some limitations, including high communication costs and catastrophic forgetting problems caused by imbalanced data distribution of devices. In this article, we design and implement IoTSL, which is an efficient distributed learning framework for efficient cloud-edge collaboration in IoT systems. IoTSL combines generative adversarial networks (GANs) and differential privacy techniques to train local data-based generators on participating devices, and generate data with privacy protection. On the one hand, IoTSL pretrains the global model using the generative data, and then fine-tunes the model using the local data to lower the communication cost. On the other hand, the generated data is used to impute the missing classes of devices to alleviate the commonly seen catastrophic forgetting phenomenon. We use three common data sets to verify the proposed framework. Extensive experimental results show that compared to the conventional SL, IoTSL significantly reduces communication costs, and efficiently alleviates the catastrophic forgetting phenomenon.

The capability of local smart sensing based on Internet of Things (IoT) devices is typically limited due to due to the inherent limitations of computational and storage capabilities. Recently, collaborative inference among multiple devices has been considered as an effective way to improve the sensing capabilities of individual IoT devices. However, the collaborative inference process still faces the challenges of data privacy leakage and inefficient collaboration. To alleviate the above issues, we design a blockchain-based collaborative inference system in this article, called CoBC, which allows each heterogeneous device node on the blockchain to customize a personalized local machine learning model according to its own hardware constraint and performance, thus improving the efficiency of resource utilization of the whole system. Meanwhile, each device node only needs to complete training locally, which significantly reduces the risk of privacy leakage due to the remote transmission of local data. CoBC improves the sensing capability of single device nodes by using collaborative inference that can obtain a more robust global inference. In addition, CoBC employs a Bayesian approximation training approach to evaluate the output uncertainty of each device node to further improve the efficiency of collaborative inference. To evaluate the performance, we deploy CoBC in a real environment and conduct a large number of simulations to evaluate the efficiency of CoBC. The simulation results demonstrate that CoBC exhibits good performance and good practicality in various criteria.

Recently we have witnessed the rise of Artificial Intelligence of Things (AIoT) and the shift of sensing paradigm from cloud-centric to the edge-centric, which effectively improves the sensing capability of intelligence transportation systems. To improve the real-time sensing performance, in this work we propose an ensemble sensing based scheme to solve the time-constraint synchronized inference problem and achieve robust inference with heterogeneous IoT devices in intelligence transportation systems. We design and implement Ensen, which incorporates various novel techniques such as customized DNN model design, KD-based model training, and dynamic deep ensemble management, etc., to achieve improved accuracy and maximize the computational resource usage of the whole sensing group. Extensive evaluations on different types of common IoT devices have shown that Ensen achieves a robust performance and can be easily extended to different types of convolutional neural networks.

As the recent rise of intelligent transportation systems (ITS), the sensing capability of vehicles has become crucial in realizing sophisticated intelligent transportation services. Collaborative sensing, an important approach to extend the sensing coverage of individual vehicles, has become an essential component of connected vehicle systems. However, due to challenges such as privacy concerns, frequent communication interruptions, customized models, and limited available data, the application of collaborative sensing in current ITS systems is still limited. In this paper, we propose BSL, a novel multi-exit split learning-based collaborative inference system. The key innovation of BSL is the introduction of multi-exit to the split network, enabling network training and collaborative inference between distributed device nodes and the cloud in a split manner. Specifically, BSL allows the device node to dynamically collaborate with the cloud by introducing the edge mode and collaboration mode, ensuring that intelligent services provided to the device will be sustained even if the communication is interrupted, which is crucial in ITS systems. We have implemented the system and evaluated it with public dataset on different embedded devices. The results demonstrate the promising performance of BSL.

Smart space has emerged as a new paradigm that combines sensing, communication, and artificial intelligence technologies to offer various customized services. A fundamental requirement of these services is person identification. Although a variety of person-identification approaches has been proposed, they suffer from several limitations in practical applications, such as low energy efficiency, accuracy degradation, and privacy issue. This article proposes an energy-harvesting-based privacy-preserving gait recognition scheme for smart space, which is named PrivGait. In PrivGait, we extract discriminative features from 1-D gait signal and design an attention-based long short-term memory (LSTM) network to classify different people. Moreover, we leverage a novel Bloom filter-based privacy-preserving technique to address the privacy leakage problem. To demonstrate the feasibility of PrivGait, we design a proof-of-concept prototype using off-the-shelf energy-harvesting hardware. Extensive evaluation results show that the proposed scheme outperforms state of the art by 6%–10% and incurs low system cost while preserving user’s privacy.

Oral health has significant impact on people’s over-all well-being. While many activity recognition systems exist in the literature, accurately sensing toothbrushing activities remains an unsolved challenging problem due to the diversity of tooth-brushing habits among different users and subtle distinctions between different brushing actions. In this work, we propose Hygiea, an energy-efficient and highly-accurate toothbrushing monitoring system which exploits IMU-based wrist-worn gesture sensing using unmodified toothbrushes. To address toothbrushing variety, Hygiea incorporates a number of novel signal preprocessing techniques to automatically transform the sensory input during arbitrary toothbrushing activities to the consistent user coordinate system. To distinguish different brushing actions, Hygiea leverages an emerging deep learning model (e.g., AT-LSTM) to achieve fine-grained activity recognitions. Moreover, a POMDP model is incorporated for sampling control to balance activity detection and energy efficiency. Extensive real-world experiments show that the Hygiea system achieves a 11.7% accuracy gain compared to the state-of-the-art while maintaining energy-efficiency and zero modification on the toothbrushes.