Ultrasound imaging has become an important modality for early breast cancer diagnosis. However, breast tumors in ultrasound images often exhibit location uncertainty, scale variations, and boundary ambiguity, which severely hinder accurate automatic classification. This paper proposes a Multiscale Convolutional and Transformer Fusion Network (MSCTFN). To address location uncertainty and boundary ambiguity, a Convolutional Block Attention Module (CBAM) is employed to process the extracted features, highlighting key tumor regions and enhancing boundary representations. To handle scale variations, a Multi-Scale Convolution (MSC) structure is designed to process features with varying receptive fields in parallel, which effectively models the multi-scale features. Additionally, a Dynamic Weighted FPN module is introduced to adaptively fuse multiscale features, significantly enhancing the model's discriminative ability and generalization performance. Experimental results on the BUSI and UDIAT breast ultrasound datasets validate the effectiveness of the proposed method, demonstrating strong generalization capability and promising clinical potential.

Post-Training Quantization (PTQ) is pivotal for deploying large language models (LLMs) within resource-limited settings by significantly reducing resource demands. However, existing PTQ strategies underperform at low bit levels (< 3 bits) due to the significant difference between the quantized and original weights. To enhance the quantization performance at low bit widths, we introduce a Mixed-precision Graph Neural PTQ (MG-PTQ) approach, employing a graph neural network (GNN) module to capture dependencies among weights and adaptively assign quantization bit-widths. Through the information propagation of the GNN module, our method more effectively captures dependencies among target weights, leading to a more accurate assessment of weight importance and optimized allocation of quantization strategies. Extensive experiments on the WikiText2 and C4 datasets demonstrate that our MG-PTQ method outperforms previous state-of-the-art PTQ method GPTQ, setting new benchmarks for quantization performance under low-bit (< 3 bits) conditions.

In the field of artificial intelligence and robotics, enabling robots to understand and execute complex tasks in unknown and open environments through natural language has become a frontier of current research. Traditional methods that rely on predefined action libraries face significant limitations when dealing with environmental diversity and task uncertainty. Accordingly, this study introduces a novel embodied intelligence framework named ‘Embodied Assistant’, which utilizes large language models (LLMs) for open-ended reasoning and adaptive task planning, can autonomously guide robots to flexibly complete challenging tasks in complex scenarios without relying on fixed action templates. By establishing effective multimodal LLM interaction pathways, this framework grants robots the ability to perceive and understand the task environment, concurrently allowing them to accurately interpret natural language instructions, and to independently plan action strategies along with robust robot trajectories. Moreover, the introduced multi-level task feedback mechanism effectively enhances the robots’ ability to self-correct and replan when they encounter planning failures. In extensive real-life testing scenarios, the proposed method achieved an 83.3% task success rate across various combined mobility and grasping tasks, significantly outperforming the most advanced baseline methods. A detailed analysis of robot fault recovery further demonstrates the substantial potential of this method in practical applications. Prompts and videos are provided at the project homepage: embodied-assistant.github.io

This article focuses on achieving efficient and safe navigation for robots in dynamic and unpredictable environments. We propose a hierarchical path planning framework that integrates global path planning with local dynamic obstacle avoidance. This framework aims to quickly plan a collision-free, shortest, and safest path for the robot while adapting the navigation path according to uncertainties in the operating environment. Global path planning is conducted using the improved gray wolf optimization algorithm (IGWO), trajectory tracking is achieved through a pure pursuit control algorithm, and a dynamic switching mechanism based on deep reinforcement learning (DRL) significantly enhances the navigation performance of the robotic system. The effectiveness of this approach has been verified through simulations and experiments. Simulation results indicate that in global path planning, the IGWO algorithm achieves faster convergence compared to algorithms, such as GWO-MP and GWO-CS. The planned path lengths are reduced by approximately 4.01% and 2.27%, respectively, and the fitness values are decreased by 4.78% and 1.9%, demonstrating superior path planning performance. For local dynamic obstacle avoidance, both single-robot and multirobot systems successfully avoided obstacles in multiple experiments. Finally, physical experiments conducted in various complex scenarios show that both single-robot and multirobot systems can effectively execute global planning and respond to unexpected obstacles. These results further demonstrate the method’s wide applicability and robust performance across diverse complex environments.

With the rapid advancement of deep learning technology, the problem of object detection in the field of remote sensing has received increasing attention. However, existing methods still face challenges when dealing with remote sensing image detection tasks, such as insufficient accuracy in small object recognition, an imbalance between recognition precision and detection efficiency, and difficulties in handling multi-scale variations. Based on the above problems, this letter proposes a lightweight and efficient remote sensing image object detector MERS-Net. Firstly, we propose an enhanced feature extraction module SCFE based on sub-channel mixing to improve the feature extraction capability of the model under complex scale transformation. Secondly, we incorporated the WaveletPool module along with the GSConv and VoVGSCSP modules from Slimneck into the model to reduce computational parameters. Finally, we designed a lightweight detection head MRF-Detect based on the parameter sharing mechanism to improve the recognition capability of the model under limited hardware conditions. We verified the effectiveness of MERS-Net on the public datasets DOTA, AI-TOD, and DIOR datasets. Compared with current mainstream algorithms such as YOLOv11, the average accuracy (mAP) is improved by 2.6%, 2.5% and 2.7%, respectively. At the same time, the parameter quantity and GFLOPs are significantly reduced, showing its superior detection performance.

Summary This article addresses the issue of ensuring model accuracy and training efficiency in a constrained federated learning environment. In an actual federated learning environment, each device's software, hardware, and network conditions are heterogeneous. Some terminal devices may not be able to undertake the work assigned by the server, resulting in poor model accuracy and slower convergence speed. However, existing research cannot ensure that each terminal device participating in training can handle the workload allocated by the system without collecting too much equipment information. This article proposes the cloud‐edge‐terminal collaborative self‐adaptive federated learning framework (CCSFLF) to solve this problem. This framework combines the advantages of federated learning and edge computing, reduces the probability that devices cannot handle the workload of system allocation, solves the system heterogeneity, and improves the efficiency of federated learning. CCSFLF can adaptively adjust the number of training tasks for terminal devices and select valuable training participants using a terminal device selection strategy. Multiple edge servers can simultaneously aggregate local models. Cloud servers are responsible for the aggregation and task distribution of global models. The above strategy enables the framework to have a faster convergence rate and higher model accuracy. The experimental results confirm that this framework can reduce the dropout rate of terminal devices by more than 5% in heterogeneous federated learning systems, improve the model accuracy by about 2%, and reduce the training time by 1/3 compared with similar methods, with better performance and applicability.

The anomaly detection of outliers and fault diagnosis in the imperfect time-series data is crucial in aero-engine industry. Early detection of the rotating stall is significant for the active stabilization control of the axial compressors, because the rotating stall and surge can cause enormous vibratory stresses in the blades and gives rise to surge to limit the performance of compressors. An aerodynamic instability inception and short-length-scale periodic anomaly prior to stall onset known as stall inception in axial compressors is observed in aero-engine. Based on deep learning theory, this paper conducts the accurate and rapid detection of stall precursors based on deep neural network. The deep dilated causal convolutional neural network combined with logistic regression(LR) named as LR-WaveNet is applied to the detection and prediction of stall inception in the imperfect time-series data of axial compressors with the rotating stall. The LR-WaveNet model can implement fast anomaly detection and stall prediction in long-time term series data. Furthermore, a single LR-WaveNet can be trained to capture and learn the time-domain statistical characteristics of many different stall inception training data with equal fidelity. The trained LR-WaveNet model can rapidly detect the occurrence of anomaly point and predict the probability of rotating stall and surge in axial compressors as an early warning signal. By comparing with the time domain analysis and the wavelet analysis, the calculation results are represented with experimental data to show the effectiveness and feasibility of the stall detection approach based on LR-WaveNet. Especially, the LR-WaveNet can detect the irregular and imperfect stall inception and capture the small fault features when the characteristics of stall inception sigals or data are not very unobvious, irregular and imperfect. Note to Practitioners—Note to Practitioners–This paper was motivated by the problem of detecting the stall and surge of axial compressor for aeroengine, but it also applies to other small fault diagnosis with imperfect data or time series prediction problem. The stall Inception is very difficult to detect for the reason of imperfect data with small fault or features. These existing detection methods of stall inception can only be applied under certain conditions, because they can only extract the pre-stall features partly. Meanwhile, the excessive number of compressor sensors is demanded in this method, so the application of these traditional methods are limited in practical situations in the stall inception detection. This paper suggests a new approach using Deep Dilated Causal Convolutional Neural Networks. The deep dilated causal convolutional neural network combined with logistic regression(LR) named as LR-WaveNet is applied to the detection and prediction of stall inception in the imperfect time-series data of axial compressors with the rotating stall. In this paper, the new approach is proved to be suitable and feasible to all the types of stall inception including the irregular stall inception with imperfect data. We then show how the deep learning method can be efficiently applied in the imperfect data situation with small fault and time series prediction. The experiments suggest that this approach is feasible. In future research, we will create the novel approach and theory for the detection of stall inception and small fault diagnostics with the imperfect data.

Iterative learning control (ILC) is an appealing method in motion control applications that can achieve the performance limit of feedforward compensation in repeating tasks. Compensating for the repetitive errors rapidly (G1) and at the same time reducing the accumulation of the nonrepetitive errors effectively (G2) are two major goals of ILC. However, in the simultaneous presence of model uncertainties and noises, most ILC methods that have been used in the application cannot fully achieve these two goals. To tackle this problem, this article develops a self-tuning nonlinear ILC method (STNILC). First, a new nonlinear ILC method (NILC) is designed in the set-membership framework based on the boundedness noise-uncertainty assumption. Then, a self-tuning algorithm is developed to determine the unknown noise-uncertainty bound that is required by NILC. Theoretical results confirm that desirable transient and steady-state convergence performances can be achieved by the proposed STNILC method. Comparative experimental results on a short-stroke stage for testing the lithographic projection lens verify the superiorities of STNILC in sufficiently achieving G1 and G2.

As one of the enabling technologies for E-Health, Internet of Medical Things (IoMT) interconnects various medical devices to collect and exchange healthcare information. To enable low-delay healthcare information processing, Mobile edge computing (MEC) can be incorporated in IoMT which can process various data in proximity of the medical devices. In this paper, we propose a Combinatorial Auction and Improved Particle Swarm Optimization based Computation Offloading Approach (CA-PSO) for e-healthcare to meet the Quality of Service (QoS) requirements of low delay and low energy consumption in healthcare monitoring. Firstly, we formulate a joint optimization problem to minimize the system cost consisting of delay and energy consumption, and transform this problem into a potential game. Secondly, we use combinatorial auction algorithm to analyze the offloading situation for different channels and servers, and combine channels and servers to simplify the original problem. Then we combine the offloading combination with improved Particle Swarm Optimization (PSO) to solve the optimal offloading strategy and server resource allocation. The simulation results show that compared with the comparison algorithm, the CA-PSO algorithm has achieved better performance in terms of average processing cost, delay, and energy consumption.

Aggressive tennis shots with high ball speed are the key factor in winning a tennis match. Today's tennis players are increasingly focused on improving ball speed. As a result, in recent tennis tournaments, records of tennis shot speeds are broken again and again. The traditional method for calculating the tennis ball speed uses multiple high-speed cameras and computer vision technology. This method is very expensive and hard to set up. Another way to calculate the tennis ball speed is to use motion sensors, which are lower cost and easier to set up. In this paper, we propose an approach for tennis ball speed estimation based on a racket-mounted motion sensor. We divide the tennis strokes into three categories: serve, groundstroke, and volley. For a serve, a regression model is proposed to estimate the ball speed. For a groundstroke or volley, two models are proposed: a regression model and a physical model. We use the physical model to estimate the ball speed for advanced players and the regression model for beginner players. Under the leave-one-subject-out cross-validation test, evaluation results show that TennisEye is 10.8% more accurate than the state-of-the-art work.

Our study reveals that facilitating warm launch of just five smartphone applications is extremely expensive, using up to 36 percent of memory. Further investigation of 20 popular applications indicates that rich multimedia applications have high heap usage and go above allowed boundaries, up to 5.63 times more heap than guaranteed by the system, and may cause crashes and erroneous behaviors. Therefore, we present iRAM, a personalized system that maintains optimal heap size limits to avoid crashes, efficiently maximizes free memory levels, and cleans low-priority processes to reduce application delays. The evaluation on memory hungry applications indicates that iRAM reduces application crashes by up to 14 percent, and reduces launch delays by up to 78.2 percent. In addition, the results confirm that iRAM increases free memory levels by up to 4.8 times. This performance gain comes with 3.5 percent of CPU overhead and 0.9 percent of power overhead.

This paper explores the correlation among three kinds of social relationships: face-to-face social relationship, online social relationship, and self-report social relationship. An experiment was carried out to collect users' three-dimensional social data: real-world mobile trace data, virtual-world online social data, and self-report social data. By analyzing network structure, we find that friendship in online social networks can better describe self-report friendship compared to friendship created by frequent physical encounters. Several supervised classifiers with the combination of features extracted from mobile trace data and online social data are used to predict the self-report social relationship under different social strengths. Results show that the proposed model can correctly predict more than 80% friends under strongest social tie strength. What is more, we define social popularity according to social relationships self-reported by users. By comparing social popularity with online and offline social behaviors, we find diversity in weekend is a good measure to describe social popularity.

Recently, the content-based publish/subscribe (pub/sub) paradigm has been gaining popularity in opportunistic mobile networks (OppNets) for its flexibility and adaptability. Since nodes in OppNets are controlled by humans, they often behave selfishly. Therefore, stimulating nodes in selfish OppNets to collect, store, and share contents efficiently is one of the key challenges. Meanwhile, guaranteeing the freshness of contents is also a big problem for content dissemination in OppNets. In this paper, in order to solve these problems, we propose an incentive-driven and freshness-aware pub/sub Content Dissemination scheme, called ConDis, for selfish OppNets. In ConDis, the Tit-For-Tat (TFT) scheme is employed to deal with selfish behaviors of nodes in OppNets. Moreover, a novel content exchange protocol is proposed when nodes are in contact. Specifically, during each contact, the exchange order is determined by the content utility, which represents the usefulness of a content for a certain node, and the objective of nodes is to maximize the utility of the content inventory stored in their buffer. Extensive realistic trace-driven simulation results show that ConDis is superior to other existing schemes in terms of total freshness value, total delivered contents, and total transmission cost.

In this paper, we develop a smart phone-based driving behavior evaluation system, named Join Driving, which helps drivers notice how aggressive their driving behaviors are and be aware of the riding comfort level of passengers. The proposed evaluation system is made of two parts: driving events detection and evaluation part and riding comfort level evaluation part. In driving events detection and evaluation part, the proposed system, Join Driving, first presents a model to detect drivers' driving events, based on the data collected from the acceleration, orientation and GPS sensors in smart phones. Then, based on the detected drivers' driving events, Join Driving implements a novel scoring mechanism to quantitatively evaluate how aggressive these driving events are. In riding comfort level evaluation part, the proposed system gives the specific scores to rate passengers' riding comfort level based on ISO 2631. Finally, several practical experiments are conducted to evaluate the effectiveness of the proposed scoring system.

In Opportunistic Mobile Networks (OppNets), a large amount of energy is consumed by idle listening, instead of infrequent data exchange. This makes energy saving a challenging and fundamental problem in OppNets, since nodes are typically battery-powered. Asynchronous duty-cycle operation is a promising approach for energy saving in OppNets, however, if its working schedule is not effectively designed, it may also cause significant network performance degradation. Therefore, it is pressing to design an energy-efficient working schedule for duty-cycle OppNets. In this paper, we first analyze the contact process in duty-cycle OppNets, then propose an adaptive working schedule for duty-cycle OppNets. The proposed schedule uses the past recorded contact histories to predict the future contact information, so as to adaptively configure the working schedule of each node in the network. Finally, extensive real trace-driven simulations are conducted to evaluate the performance of our proposed schedule. The results show that our proposed adaptive working schedule is superior to the random working schedule in terms of delivery ratio and delivery delay.

Recently, the content-based publish/subscribe (pub/sub) paradigm is gaining popularity in opportunistic mobile networks (OppNets) for its flexibility and adaptability. Since nodes in OppNets is controlled by humans, they often behave selfishly with an aim to maximize their own revenues without considering the performance of others. Therefore, stimulating nodes in OppNets to collect, store, and share content efficiently is one of the key challenges under this scheme. Meanwhile, guaranteeing the freshness of content is also a big problem for content dissemination in OppNets. In this paper, in order to solve these problems, we propose an incentive-driven and freshness-aware pub/sub content dissemination scheme, called ConDis (Content Dissemination), for selfish OppNets. In ConDis, the Tit-For-Tat (TFT) scheme is employed to deal with selfish behaviors of nodes in OppNets. ConDis also implements a novel content exchange protocol when nodes are in contact. Specifically, during each contact, the exchange order is determined by the content utility, which is calculated by the direct subscribed value and the indirect subscribed value, and the objective of nodes is to maximize the utility of the content inventory stored in their buffer. Extensive realistic trace-driven simulation results show that ConDis is superior to other existing schemes in terms of total freshness value, total delivered contents, and total transmission cost.

In Opportunistic Mobile Networks (OppNets), a large amount of energy is consumed by idle listening, instead of infrequent data exchanging. This makes energy saving become a serious problem in OppNets, as nodes are typically with limited energy supplies. Duty-cycle operation is a promising approach for energy saving in OppNets, however, it may cause the degradation of network connectivity. Therefore, there exists a tradeoff relationship between energy saving and network connectivity in duty-cycle OppNets. In this paper, we propose a model to quantify the contact probability (a metric of connectivity) in duty-cycle OppNets based on the realistic mobility trace. Then the tradeoff between energy saving and the contact probability under different situations is derived and analyzed. Our results show that the duty-cycled nodes can guarantee an energy saving of around 50% without impacting the contact probability if the period meets a certain condition, and higher energy saving can be achieved at the cost of reducing the contact probability. Finally, realistic trace-driven simulations are performed to validate the correctness of our results.