Search Results (19,702)

In the field of embedded and edge devices, efforts have been made to make deep neural network models smaller due to the limited size of the available memory and the low computational efficiency. Typical model footprints are under 100 KB. However, for some applications, models of this size are too large. In low-voltage sensors, signals must be processed, classified or predicted with an order of magnitude smaller memory. Model downsizing can be performed by limiting the number of model parameters or quantizing their weights. These types of operations have a negative impact on the accuracy of the deep network. This study tested the effect of model downscaling techniques on accuracy. The main idea was to reduce neural network models to 3 k parameters or less. Tests were conducted on three different neural network architectures in the context of three separate research problems, modeling real tasks for small networks. The impact of the reduction in the accuracy of the network depends mainly on its initial size. For a network reduced from 40 k parameters, a decrease in accuracy of 16 percentage points was achieved, and for a network with 20 k parameters, a decrease of 8 points was achieved. To obtain the best results, knowledge distillation and quantization-aware training methods were used for training. Thanks to this, the accuracy of the 4-bit networks did not differ significantly from the 8-bit ones and their results were approximately four percentage points worse than those of the full precision networks. For the fully connected network, synthesis to ASIC (application-specific integrated circuit) was also performed to demonstrate the reduction in the silicon area occupied by the model. The 4-bit quantization limits the silicon area footprint by 90%. Full article

Diabetes is a chronic condition, and traditional monitoring methods are invasive, significantly reducing the quality of life of the patients. This study proposes the design of an innovative system based on a microcontroller that performs real-time ECG acquisition and evaluates the presence of diabetes using an Edge-AI solution. A spectrogram-based preprocessing method is combined with a 1-Dimensional Convolutional Neural Network (1D-CNN) to analyze the ECG signals directly on the device. By applying quantization as an optimization technique, the model effectively balances memory usage and accuracy, achieving an accuracy of 89.52% with an average precision and recall of 0.91 and 0.90, respectively. These results were obtained with a minimal memory footprint of 347 kB flash and 23 kB RAM, showcasing the system’s suitability for wearable embedded devices. Furthermore, a custom PCB was developed to validate the system in a real-world scenario. The hardware integrates high-performance electronics with low power consumption, demonstrating the feasibility of deploying Edge-AI for non-invasive, real-time diabetes detection in resource-constrained environments. This design represents a significant step forward in improving the accessibility and practicality of diabetes monitoring. Full article

The Internet of Things (IoT) contains many devices that can compute and communicate, creating large networks. Industrial Internet of Things (IIoT) represents a developed application of IoT, connecting with embedded technologies in production in industrial operational settings to offer sophisticated automation and real-time decisions. Still, IIoT compels significant cybersecurity threats beyond jamming and spoofing, which could ruin the critical infrastructure. Developing a robust Intrusion Detection System (IDS) addresses the challenges and vulnerabilities present in these systems. Traditional IDS methods have achieved high detection accuracy but need improved scalability and privacy issues from large datasets. This paper proposes a Fog-enabled Federated Learning-based Intrusion Detection System (FFL-IDS) utilizing Convolutional Neural Network (CNN) that mitigates these limitations. This framework allows multiple parties in IIoT networks to train deep learning models with data privacy preserved and low-latency detection ensured using fog computing. The proposed FFL-IDS is validated on two datasets, namely the Edge-IIoTset, explicitly tailored to environments with IIoT, and CIC-IDS2017, comprising various network scenarios. On the Edge-IIoTset dataset, it achieved 93.4% accuracy, 91.6% recall, 88% precision, 87% F1 score, and 87% specificity for jamming and spoofing attacks. The system showed better robustness on the CIC-IDS2017 dataset, achieving 95.8% accuracy, 94.9% precision, 94% recall, 93% F1 score, and 93% specificity. These results establish the proposed framework as a scalable, privacy-preserving, high-performance solution for securing IIoT networks against sophisticated cyber threats across diverse environments. Full article

The failure of different chips under working conditions is influenced by various stress states such as different voltages, temperatures, stress durations, and stress variations. Therefore, the failure time has a great degree of dispersion, and similar chips may exhibit different failure mechanisms due to variations in their working environments. This paper proposes three system-on-chip reliability failure prediction unit circuits: the time-dependent dielectric breakdown prediction circuit, the negative bias temperature instability prediction circuit, and the hot carrier injection prediction circuit. These circuits are embedded within the main chip, enabling real-time failure prediction and reliability mechanism diagnosis in the same working environment as the main chip. The three reliability failure prediction circuits are compact and energy efficient, allowing for their integration into a system on a chip as IP cores that provide early warning signals before system-on-chip failure. Compared to traditional reliability prediction methods, this approach offers the advantages of accurately identifying failure mechanisms, predicting failure times, and enabling real-time online monitoring. Full article

The chicken is the world’s most farmed animal. In this work, we introduce the Chicks4FreeID dataset, the first publicly available dataset focused on the reidentification of individual chickens. We begin by providing a comprehensive overview of the existing animal reidentification datasets. Next, we conduct closed-set reidentification experiments on the introduced dataset, using transformer-based feature extractors in combination with two different classifiers. We evaluate performance across domain transfer, supervised, and one-shot learning scenarios. The results demonstrate that transfer learning is particularly effective with limited data, and training from scratch is not necessarily advantageous even when sufficient data are available. Among the evaluated models, the vision transformer paired with a linear classifier achieves the highest performance, with a mean average precision of 97.0%, a top-1 accuracy of 95.1%, and a top-5 accuracy of 100.0%. Our evaluation suggests that the vision transformer architecture produces higher-quality embedding clusters than the Swin transformer architecture. All data and code are publicly shared under a CC BY 4.0 license. Full article

Background: Marginal adaptation is one of biggest challenges in restorative dentistry, mainly in restoration of composite resin. Even with development of lower shrinkage materials, like those which use the polymer silorane, it is necessary a faithful clinical protocol in order to obtain better results and more clinical durability of restorative procedure. The purpose of this study was investigated if different polymers can influence on marginal adaptation of different composite resin restorations. Material and methods: Twenty class V cavities were confectioned in twenty human molars in a standardized way. Ten dentists received one tooth each and a questionnaire to describe their own clinical protocol. The other 10 teeth were restored by one researcher as control group. The cavities received the adhesive system than increments of composite resin, and so the restorations were finished with finishing bur (KG Sorensen, São Paulo, SP, Brazil) to remove the excess, followed by sof-lex discs (3M ESPE). After this stage, all of specimens were embedded in 1% Methylene blue (Prolabo, Paris, France) during 48 h. The evaluation of pigment penetration into the interfaces was performed after specimens were washed in distilled water and longitudinally sectioned with steel Diamond disc in low velocity. Results: There was significant statistical difference between the different techniques using compose resin. Conclusions: it was possible conclude that the clinical protocol to perform dental restoration interfere dramatically in the final results of restorative procedure. Full article

With the flourishing of social media platforms, data in social networks, especially user-generated content, are growing rapidly, which makes it hard for users to select relevant content from the overloaded data. Recommender systems are thus developed to filter user-relevant content for better user experiences and also the commercial needs of social platform providers. Graph neural networks have been widely applied in recommender systems for better recommendation based on past interactions between users and corresponding items due to the graph structure of social data. Users might also be influenced by their social connections, which is the focus of social recommendation. Most works on recommendation systems try to obtain better representations of user embeddings and item embeddings. Compared with recommendation systems only focusing on interaction graphs, social recommendation has an additional task of combining user embedding from the social graph and interaction graph. This paper proposes a new method called SocialJGCF to address these problems, which applies Jacobi-Polynomial-Based Graph Collaborative Filtering (JGCF) to the propagation of the interaction graph and social graph, and a graph fusion is used to combine the user embeddings from the interaction graph and social graph. Experiments are conducted on two real-world datasets, epinions and LastFM. The result shows that SocialJGCF has great potential in social recommendation, especially for cold-start problems. Full article

This study presents a detailed characterization of iterative steganography, a unique class of information-hiding techniques, and proposes a formal mathematical model for their description. A novel quantitative measure, the Incremental Information Function (IIF), is introduced to evaluate the process of information gain in iterative steganographic methods. The IIF offers a comprehensive framework for analyzing the step-by-step process of embedding information into a cover medium, focusing on the cumulative effects of each iteration in the encoding and decoding cycles. The practical application and efficacy of the proposed method are demonstrated using detailed case studies in video steganography. These examples highlight the utility of the IIF in delineating the properties and characteristics of iterative steganographic techniques. The findings reveal that the IIF effectively captures the incremental nature of information embedding and serves as a valuable tool for assessing the robustness and capacity of steganographic systems. This research provides significant insights into the field of information hiding, particularly in the development and evaluation of advanced steganographic methods. The IIF emerges as an innovative and practical analytical tool for researchers, offering a quantitative approach to understanding and optimizing iterative steganographic techniques. Full article

In successful peer discussions students respond to each other and benefit from supports that focus discussion on one another’s ideas. We explore using artificial intelligence (AI) to form groups and guide peer discussion for grade 7 students. We use natural language processing (NLP) to identify student ideas in science explanations. The identified ideas, along with Knowledge Integration (KI) pedagogy, informed the design of a question bank to support students during the discussion. We compare groups formed by maximizing the variety of ideas among participants to randomly formed groups. We embedded the chat tool in an earth science unit and tested it in two classrooms at the same school. We report on the accuracy of the NLP idea detection, the impact of maximized versus random grouping, and the role of the question bank in focusing the discussion on student ideas. We found that the similarity of student ideas limited the value of maximizing idea variety and that the question bank facilitated students’ use of knowledge integration processes. Full article

To address the difficulty in detecting workers’ violation behaviors in electric power construction scenarios, this paper proposes an innovative method that integrates knowledge reasoning and progressive multi-level distillation techniques. First, standards, norms, and guidelines in the field of electric power construction are collected to build a comprehensive knowledge graph, aiming to provide accurate knowledge representation and normative analysis. Then, the knowledge graph is combined with the object-detection model in the form of triplets, where detected objects and their interactions are represented as subject–predicate–object relationship. These triplets are embedded into the model using an adaptive connection network, which dynamically weights the relevance of external knowledge to enhance detection accuracy. Furthermore, to enhance the model’s performance, the paper designs a progressive multi-level distillation strategy. On one hand, knowledge transfer is conducted at the object level, region level, and global level, significantly reducing the loss of contextual information during distillation. On the other hand, two teacher models of different scales are introduced, employing a two-stage distillation strategy where the advanced teacher guides the primary teacher in the first stage, and the primary teacher subsequently distills this knowledge to the student model in the second stage, effectively bridging the scale differences between the teacher and student models. Experimental results demonstrate that under the proposed method, the model size is reduced from 14.5 MB to 3.8 MB, and the floating-point operations (FLOPs) are reduced from 15.8 GFLOPs to 5.9 GFLOPs. Despite these optimizations, the AP50 reaches 92.4%, showing a 1.8% improvement compared to the original model. These results highlight the method’s effectiveness in accurately detecting workers’ violation behaviors, providing a quantitative basis for its superiority and offering a novel approach for safety management and monitoring at construction sites. Full article

With the widespread construction of the subway in the Chinese mainland, the environmental vibration caused by subway operation has attracted increasing attention. Train-induced environmental vibrations can cause structural deformation, uneven settlement of line foundations, and tunnel leakage, affecting the structural safety of lines and foundations. This research focuses on a segment of the Nanchang Metro Line 3, which has been chosen as the subject of investigation. A numerical model was developed to analyze the subway train-induced environmental vibration, employing the finite element method (FEM). Utilizing a numerical model, an investigation was conducted to examine the impact of train speed on the subway train-induced environmental vibration, the train-induced environmental vibration transmission characteristics were analyzed, and the control effects of vibration reduction tracks on train-induced environmental vibration were discussed. Train-induced vibration tests were also conducted on Nanchang Metro Line 3 to verify the control effects of various vibration reduction tracks. The results indicate that the subway train-induced environmental vibration rises as the train speed goes up, and the vibration peaks always appear around 63 Hz. When the train speed doubles, the Z-vibration level increases from about 5.1 dB to 5.9 dB. Subway train-induced environmental vibration shows a fluctuating decreasing trend with increasing distance from the centerline of the tunnel. The Z-vibration level reaches its maximum 4 m away from the centerline of the tunnel. Compared with the embedded sleeper, the vibration-damping fastener exhibits a vibration reduction effect of about 9 dB to 18 dB, the rubber vibration-damping pad exhibits a better vibration reduction effect of about 16 dB to 24 dB, and the steel spring floating plate exhibits the best vibration-damping effect of about 18 dB to 28 dB. The calculated Z-vibration levels are basically consistent with the measured values, indicating the accuracy of the calculated results of the control effects of the vibration reduction tracks. Full article

The controlled formation and stabilization of nanoparticles is of fundamental relevance for materials science and key to many modern technologies. Common synthetic strategies to arrest growth at small sizes and prevent undesired particle agglomeration often rely on the use of organic additives and require non-aqueous media and/or high temperatures, all of which appear critical with respect to production costs, safety, and sustainability. In the present work, we demonstrate a simple one-pot process in water under ambient conditions that can produce particles of various transition metal carbonates and sulfides with sizes of only a few nanometers embedded in a silica shell, similar to particles derived from more elaborate synthesis routes, like the sol–gel process. To this end, solutions of soluble salts of metal cations (e.g., chlorides) and the respective anions (e.g., sodium carbonate or sulfide) are mixed in the presence of different amounts of sodium silicate at elevated pH levels. Upon mixing, metal carbonate/sulfide particles nucleate, and their subsequent growth causes a sensible decrease of pH in the vicinity. Dissolved silicate species respond to this local acidification by condensation reactions, which eventually lead to the formation of amorphous silica layers that encapsulate the metal carbonate/sulfide cores and, thus, effectively inhibit any further growth. The as-obtained carbonate nanodots can readily be converted into the corresponding metal oxides by secondary thermal treatment, during which their nanometric size is maintained. Although the described method clearly requires optimization towards actual applications, the results of this study highlight the potential of bottom-up self-assembly for the synthesis of functional nanoparticles at mild conditions. Full article

Strain tolerance is a crucial factor affecting the thermal life of coatings, and a higher strain tolerance can effectively alleviate the thermal stresses on coatings during thermal shock. To improve the strain tolerance, the coating structure was optimized by introducing an intermediate transition layer in this study. The intermediate transition layer material was prepared using a 1:1 volume ratio mixture of 6–8 wt. % Yttria-stabilized zirconia (YSZ) and NiCrAlY powders in the experiments. The coating structure consisted of an Al₂O₃-GdAlO₃ (AGAP) anti-erosion layer, a YSZ layer, an intermediate transition layer, and a bonding layer from top to bottom. After thermal shock experiments at 1500 °C, the coatings with the addition of the intermediate transition layer exhibited different failure modes, with the crack location shifting from between the YSZ and the bonding layer to within the intermediate transition layer, compared to the coatings without the intermediate transition layer. Finite element simulation analysis showed that the intermediate transition layer effectively increased the strain tolerance of the coating and significantly reduced the thermal stress. Furthermore, incorporating an embedded micron agglomerated particle-based (EMAP) thermal barrier coating structure into the intermediate transition layer effectively alleviated thermal stresses and enhanced the coating’s thermal insulation performance. Full article

Herein, biochars derived from corn stalks, rice husks, and bamboo powder were modified by nitric acid oxidation and sodium hydroxide alkali activation to identify efficient and cost-effective polycyclic aromatic hydrocarbon-adsorbent and microbial-immobilized carriers. The surface characterization and adsorption investigation results suggested that acid/alkali modification promoted the phenanthrene removal ability in an aqueous solution of biochars via facilitating π–π/n–π electron donor–acceptor interactions, electrostatic interactions, hydrogen bonds, and hydrophobic interactions. Subsequently, the degrading bacteria Rhodococcus sp. DG1 was successfully immobilized on the rice husk-derived biochar with nitric acid oxidation (RBO), which exhibited the maximum phenanthrene adsorption efficiency (3818.99 µg·g⁻¹), abundant surface functional groups, and a larger specific surface area (182.6 m²·g⁻¹) and pore volume (0.141 m³·g⁻¹). Degradation studies revealed that the microorganisms immobilized on RBO by the adsorption method yielded a significant phenanthrene removal rate of 80.15% after 30 days, which was 38.78% higher than that of the control. Conversely, the polymer gel network-based microenvironment in the microorganism-immobilized RBO by the combined adsorption–embedding method restricted the migration and diffusion of nutrients and pollutants in the reaction system. This study thus introduces an innovative modified biochar-based microbial immobilization technology characterized by a simple design, convenient operation, and high adsorption efficiency, offering valuable insights into material selection for PAH contamination bioremediation. Full article

Geographic knowledge graph representation learning embeds entities and relationships in geographic knowledge graphs into a low-dimensional continuous vector space, which serves as a basic method that bridges geographic knowledge graphs and geographic applications. Previous geographic knowledge graph representation methods primarily learn the vectors of entities and their relationships from their spatial attributes and relationships, which ignores various semantics of entities, resulting in poor embeddings on geographic knowledge graphs. This study proposes a two-stage multimodal geographic knowledge graph representation (MGKGR) model that integrates multiple kinds of semantics to improve the embedding learning of geographic knowledge graph representation. Specifically, in the first stage, a spatial feature fusion method for modality enhancement is proposed to combine the structural features of geographic knowledge graphs with two modal semantic features. In the second stage, a multi-level modality feature fusion method is proposed to integrate heterogeneous features from different modalities. By fusing the semantics of text and images, the performance of geographic knowledge graph representation is improved, providing accurate representations for downstream geographic intelligence tasks. Extensive experiments on two datasets show that the proposed MGKGR model outperforms the baselines. Moreover, the results demonstrate that integrating textual and image data into geographic knowledge graphs can effectively enhance the model’s performance. Full article