Search

To extract knowledge from the large data collected by edge devices, a traditional cloud-based approach that requires data upload may not be feasible due to communication bandwidth limitations as well as privacy and security concerns of end-users. A novel privacy-preserving edge intelligent computing framework for image classification in IoT is proposed to address these challenges. Specifically, the autoencoder will be trained unsupervised at each edge device individually, and then the obtained latent vectors transmitted to the edge server for the training of a classifier. This framework would reduce the communication overhead and protect end-users’ data. Compared to federated learning, the training of the classifier in the proposed framework is not subject to the constraints of the edge devices, and the autoencoder can be trained independently at each edge device without any server involvement. Compared to collaborative intelligence such as SplitNN, the proposed method does not suffer from high communication cost as noticed in SplitNN. Furthermore, the privacy of the end-users’ data is protected by transmitting latent vectors and without the additional cost of encryption. Experimental results provide insights on the image classification performance vs. various design parameters such as the data compression ratio of the autoencoder and the model complexity.

Dapagliflozin (DAPA) exerts cardioprotective effects in non-diabetic patients. Nonetheless, the protective mechanism remains unknown. This study aims to evaluate the performance of DAPA on cardiac function and remodeling as well as its potential mechanism in mice with myocardial infarction (MI). Here, a MI mice model was established. One week after MI, mice were treated with saline or DAPA (1.5 mg/kg/day) for 4 weeks. At the end of this study, echocardiography was performed to assess cardiac structure and function. Myocardial apoptosis was analyzed by Western blot and immunofluorescence. Inflammatory cytokines and cardiac fibrosis were analyzed by real-time PCR and Masson's trichrome stain, respectively. Results showed that DAPA improved cardiac structure and function, attenuated cardiac fibrosis, and inhibited inflammatory cytokines and myocardial apoptosis. Moreover, the inhibition of PI3K/AKT/mTOR pathway might be related to the cardioprotective role of DAPA. These findings reveal that dapagliflozin is a potential therapeutic agent for MI patients without diabetes.

Background. Global longitudinal strain (GLS) can be assessed by speckle tracking echocardiography (STE) to express the degree of cardiac fibrosis. Qianyangyuyin (QYYY) granules can effectively improve GLS in hypertensive patients. Using a hypertensive rat model, we carried out speckle tracking echocardiography to validate the effect of QYYY in diminishing LV remodeling. Methods. We randomly divided 16 spontaneously hypertensive rats (SHRs) into SHR, SHR + valsartan (SHR + V), SHR + low-dose QYYY (SHR + QL), and SHR + high-dose QYYY (SHR + QH) groups, with four rats in each group. Another group of 4 Wistar-Kyoto (WKY) rats were selected into a normal control (WKY) group. At the 8th week, conventional echocardiographic parameters were measured by GE Vivid E95 ultrasound (12S probe, 10–12 MHz) and GLS by speckle tracking echocardiography with EchoPAC (version 203) software. HE and Masson’s trichrome staining were performed to detect the cardiomyocyte width and collagen volume fraction after rat sacrifice. Collagen I, α-SMA, S100A4, TGF-β, Smad 3, MYH6, and MYH7 were further analyzed by Western blot. Results. The absolute values of GLS significantly increased in the SHR + QH group compared to the SHR group, while the CVF and CW values significantly decreased. In addition, Collagen I, α-SMA, S100A4, TGF-β, Smad3, MYH7, and MYH7/MYH6 ratio remarkably reduced in the SHR + QH group. The value of GLS could be repetitively measured and positively correlated with the collagen volume fraction of the myocardium and the cardiomyocyte width of the left ventricular free wall. Conclusions. GLS is a reliable indicator to evaluate the therapeutic effect on left ventricular remodeling in hypertension. QYYY granules can inhibit the development of cardiac fibrosis in the hypertensive rat model.

BACKGROUND: Perfluoropropane droplets formulated from commercial microbubbles exhibit different acoustic characteristics than their parent microbubbles, most likely from enhanced endothelial permeability. This enhanced permeability may permit delayed echo-enhancement imaging (DEEI) similar to delayed enhancement magnetic resonance imaging (DE-MRI). We hypothesized this would allow detection and quantification of myocardial scar.METHODS: In 15 pigs undergoing 90 minutes of left anterior descending ischemia by either balloon (n = 13) or thrombotic occlusion (n = 2), DE-MRI was performed at 2-24 days postocclusion. Delayed echo-enhancement imaging was performed at 2-4 minutes following an intravenous injection of 1 mL of 50% Definity (Lantheus Medical) compressed into 180 nm droplets; DEEI was attempted in all pigs with single-pulse harmonic imaging at 1.7 transmit/3.4 MHz receive. Myocardial defects observed with DEEI were quantified (percentage of infarct area) and compared with DE-MRI as well as postmortem staining. In six pigs, multipulse low-mechanical index (MI) fundamental nonlinear imaging (FNLI) with intermittent high-MI impulses was performed to determine whether droplet activation within the infarct zone was achievable with a longer pulse duration.RESULTS: The range of infarct size area by DE-MRI ranged from 0% to 46% of total left ventricular area. Single-pulse harmonic imaging detected a contrast defect that correlated closely with infarct area by DE-MRI (r = 0.81, P = .0001). The FNLI high-MI impulses resulted in droplet activation in both the infarct and normal zones. Harmonic subtraction of the FNLI images resulted in infarct zone enhancement that also correlated closely with infarct size (r = 0.83; P = .04). Droplets were observed on postmortem transmission electron microscopy within myocytes of the infarct and remote normal zone.CONCLUSION: Intravenously Definity nanodroplets can be utilized to detect and quantify infarct zone at the bedside using DEEI techniques.

Nanoscale-diameter liquid droplets from commercially available microbubbles may optimize thrombus permeation and subsequent thrombus dissolution (TD). Thrombi were made using fresh porcine arterial whole blood and placed in an in vitro vascular simulation. A diagnostic ultrasound probe in contact with a tissue-mimicking phantom tested intermittent high-mechanical-index (HMI) fundamental multipulse (focused ultrasound [FUS], 1.8 MHz) versus harmonic single-pulse (HUS, 1.3 MHz) modes during a 10-min infusion of Definity nanodroplets (DNDs), Definity microbubbles (DMBs) or saline. The ability of FUS and intravenous DNDs to improve epicardial and microvascular flow was then tested in four pigs with left anterior descending thrombotic occlusion. Sixty in vitro thrombi were tested, 20 in each group. Percentage TD was significantly higher for DND-treated thrombi than DMB-treated thrombi and controls (DNDs: 42.4%, DMBs: 26.7%, saline: 15.0%; p < 0.0001 vs. control). The highest %TD was seen in the HMI FUS-treated DND group (51 ± 17% TD). HMI FUS detected droplet activation within the risk area in three of four pigs with left anterior descending thrombotic occlusion and re-canalized the epicardial vessel in two. DNDs with intermittent diagnostic HMI ultrasound resulted in significantly more intravascular TD than DMBs and have potential for coronary and risk area thrombolysis.

The empowering unmanned aerial vehicles (UAVs) have been extensively used in providing intelligence such as target tracking. In our field experiments, a pretrained convolutional neural network (CNN) is deployed at UAV to identify a target (a vehicle) from the captured video frames and enable the UAV to keep tracking. However, this kind of visual target tracking demands a lot of computational resources due to the desired high inference accuracy and stringent delay requirement. This motivates us to consider offloading this type of deep learning (DL) tasks to a mobile-edge computing (MEC) server due to the limited computational resource and energy budget of the UAV and further improve the inference accuracy. Specifically, we propose a novel hierarchical DL tasks distribution framework, where the UAV is embedded with lower layers of the pretrained CNN model while the MEC server (MES) with rich computing resources will handle the higher layers of the CNN model. An optimization problem is formulated to minimize the weighted-sum cost, including the tracking delay and energy consumption introduced by communication and computing of UAVs while taking into account the quality of data (e.g., video frames) input to the DL model and the inference errors. Analytical results are obtained and insights are provided to understand the tradeoff between the weighted-sum cost and inference error rate in the proposed framework. Numerical results demonstrate the effectiveness of the proposed offloading framework.

In a level-5 autonomous driving system, the autonomous driving vehicles (AVs) are expected to sense the surroundings via analyzing a large amount of data captured by a variety of onboard sensors in near-real-time. As a result, enormous computing costs will be introduced to the AVs for processing the tasks with the deployed machine learning (ML) model, while the inference accuracy may not be guaranteed. In this context, the advent of edge intelligence (EI) and sixth-generation (6G) wireless networking are expected to pave the way to more reliable and safer autonomous driving by providing multi-access edge computing (MEC) together with ML to AVs in close proximity. To realize this goal, we propose a two-tier EI-empowered autonomous driving framework. In the autonomous-vehicles tier, the autonomous vehicles are deployed with the shallow layers by splitting the trained deep neural network model. In the edge-intelligence tier, an edge server is implemented with the remaining layers (also deep layers) and an appropriately trained multi-task learning (MTL) model. In particular, obtaining the optimal offloading strategy (including the binary offloading decision and the computational resources allocation) can be formulated as a mixed-integer nonlinear programming (MINLP) problem, which is solved via MTL in near-real-time with high accuracy. On another note, an edge-vehicle joint inference is proposed through neural network segmentation to achieve efficient online inference with data privacy-preserving and less communication delay. Experiments demonstrate the effectiveness of the proposed framework, and open research topics are finally listed.

Intermittent high-mechanical index (MI) impulses from a transthoracic ultrasound transducer are recommended for regional wall motion analysis and assessment of myocardial perfusion following intravenous administration of ultrasound enhancing agents (UEAs). High-MI impulses (1.0) applied in this setting have also been shown to increase microvascular blood flow through a purinergic signaling pathway, but their effects on left ventricular (LV) myocardial function are unknown. Therefore, the aim of this study was to investigate the effect of transthoracic intermittent high-MI impulses during intravenous UEA infusion in patients with normal and abnormal resting systolic function.Fifty patients referred for echocardiography to evaluate LV systolic function during continuous infusion of UEAs (Definity 3% infusion) were prospectively assigned to low-MI (0.2) imaging alone (group 1) or low-MI (0.2) imaging with intermittent high-MI impulses (five frames, 1.8 MHz, MI = 1.0-1.2) applied at least two times in each apical window to clear myocardial contrast (group 2). Global longitudinal strain (GLS) measurements were obtained at baseline before UEA administration and at 5-min intervals up to 10-min after infusion completion.There were no differences between groups with respect to age, gender, resting GLS, biplane LV ejection fraction, or cardiac risk factors. Resting GLS in group 1 was -15.5 ± 5.2% before UEA infusion and -15.5 ± 5.4% at 10 min after UEA infusion. In comparison, GLS increased in group 2 (-15.3 ± 5.0 before infusion and -16.8 ± 4.8% at 10 min, P .00001). Improvements in GLS were seen in patients with normal and abnormal systolic function. Regional analysis demonstrated that the increase in strain in patients with abnormal LV ejection fractions was primarily in the apical segments (-12.0 ± 2.7% before infusion and -13.4 ± 3.4% at 10 min, P = .001).High-MI impulses during infusion of a commercially available contrast agent can improve LV systolic function and may have therapeutic effect in patients with LV dysfunction.

News in social media, such as Twitter, has been generated in high volume and speed. However, very few of them are labeled (as fake or true news) by professionals in near real time. In order to achieve timely detection of fake news in social media, a novel framework of two-path deep semisupervised learning (SSL) is proposed where one path is for supervised learning and the other is for unsupervised learning. The supervised learning path learns on the limited amount of labeled data, while the unsupervised learning path is able to learn on a huge amount of unlabeled data. Furthermore, these two paths implemented with convolutional neural networks (CNNs) are jointly optimized to complete SSL. In addition, we build a shared CNN to extract the low-level features on both labeled data and unlabeled data to feed them into these two paths. To verify this framework, we implement a Word CNN-based SSL model and test it on two data sets: LIAR and PHEME. Experimental results demonstrate that the model built on the proposed framework can recognize fake news effectively with very few labeled data.

Purpose Volumetric pancreas segmentation can be used in the diagnosis of pancreatic diseases, the research about diabetes and surgical planning. Since manual delineation is time-consuming and laborious, we develop a deep learning-based framework for automatic pancreas segmentation in three dimensional (3D) medical images. Methods A two-stage framework is designed for automatic pancreas delineation. In the localization stage, a Square Root Dice loss is developed to handle the trade-off between sensitivity and specificity. In refinement stage, a novel 2.5D slice interaction network with slice correlation module is proposed to capture the non-local cross-slice information at multiple feature levels. Also a self-supervised learning-based pre-training method, slice shuffle, is designed to encourage the inter-slice communication. To further improve the accuracy and robustness, ensemble learning and a recurrent refinement process are adopted in the segmentation flow. Results The segmentation technique is validated in a public dataset (NIH Pancreas-CT) with 82 abdominal contrast-enhanced 3D CT scans. Fourfold cross-validation is performed to assess the capability and robustness of our method. The dice similarity coefficient, sensitivity, and specificity of our results are 86.21 ± 4.37%, 87.49 ± 6.38% and 85.11 ± 6.49% respectively, which is the state-of-the-art performance in this dataset. Conclusions We proposed an automatic pancreas segmentation framework and validate in an open dataset. It is found that 2.5D network benefits from multi-level slice interaction and suitable self-supervised learning method for pre-training can boost the performance of neural network. This technique could provide new image findings for the routine diagnosis of pancreatic disease.

This letter proposes an edge learning-based offloading framework for autonomous driving, where the deep learning tasks can be offloaded to the edge server to improve the inference accuracy while meeting the latency constraint. Since the delay and the inference accuracy are incurred by wireless communications and computing, an optimization problem is formulated to maximize the inference accuracy subject to the offloading probability, the pre-braking probability, and data quality. Simulations demonstrate the superiority of the proposed offloading framework.

Long-term evolution (LTE) deployment in the unlicensed band has been a promising solution to handle the ever-increasing data traffic growth. However, spectrum sharing on unlicensed band poses a significant challenge regarding the interaction between LTE licensed assisted access (LAA) and Wi-Fi. In this paper, a radio access technology (RAT) heterogeneous network that consists of an LAA tier and a Wi-Fi tier is constructed, to achieve the coexistence and alleviate the intra-RAT and inter-RAT interference, a listen-and-talk (LAT) scheme is utilized in Wi-Fi while LAA adopts the listen-before-talk (LBT). By leveraging the full-duplex (FD) techniques in Wi-Fi, the collision can be avoided and the utilization of unlicensed spectrum can be improved. Furthermore, the sensing errors are derived based on the FD strategy and an enhanced Markov model is presented to analyze the performance of the heterogeneous network with consideration of the residual self-interference (RSI). The fairness between LAA and Wi-Fi is also investigated. At last, to ensure the performance of Wi-Fi when serious RSI exists, a switched MAC (S-MAC) that can adaptively switch between the FD mode and half-duplex (HD) mode is presented.

In this article, we propose a novel framework of mobile edge computing (MEC)-based hierarchical machine learning (ML) tasks distribution for the Industrial Internet of Things. It is assumed that a batch of ML tasks, such as anomaly detection, need to be executed timely in an MEC setting, where the devices have limited computing capability while the MEC server (MES) has rich computing resources. Thus, a small ML model for the device and a deep ML model for the MES are pretrained offline using historical data, and then they are deployed accordingly. However, offloading tasks to the MES introduces communications delay. Thus, each device must decide the portion of the tasks to offload to minimize the processing delay. Since the delay and the error of data processing are incurred by communications and ML computing, a joint optimization problem is formulated to minimize the total delay subject to the ML model complexity and inference error rate, data quality, computing capability at the device and MES, and communications bandwidth. A closed-form solution is derived analytically and an optimal offloading strategy selection algorithm is proposed. Insights are provided to understand the tradeoff between communications and ML computing in offloading decisions, and the effects of key parameters in the proposed algorithm are investigated. The numerical results demonstrate the effectiveness of the proposed algorithm.

Through machine learning, this paper changes the fundamental assumption of the traditional medium access control (MAC) layer design. It obtains the capability of retrieving the information even the packets collide by training a deep neural network offline with the historical radio frequency (RF) traces and inferring the STAs involved collisions online in near-real-time. Specifically, we propose a MAC protocol based on intelligent spectrum learning for the future wireless local area networks (WLANs), called SL-MAC. In the proposed MAC, an access point (AP) is installed with a pre-trained convolutional neural network (CNN) model to identify the stations (STAs) involved in the collisions. In contrast to the conventional contention-based random medium access methods, e.g., IEEE 802.11 distributed coordination function (DCF), the proposed SL-MAC protocol seeks to schedule data transmissions from the STAs suffering from the collisions. To achieve this goal, we develop a two-step offline training algorithm enabling the AP to sense the spectrum with the aid of the CNN. In particular, on receiving the overlapped signal(s), the AP firstly predicts the number of STAs involving collisions and then further identifies the STAs’ ID. Furthermore, we analyze the upper bound of throughput gain brought by the CNN predictor and investigate the impact of the inference error on the achieved throughput. Extensive simulations show the superiority of the proposed SL-MAC and allow us to gain insights on the trade-off between performance gain and the inference accuracy.

Intelligent capabilities are of utmost importance in future wireless communication systems. For optimum resource utilization, wireless communication systems require knowledge of the prevalent situation in a frequency band through learning. To learn appropriately, it is imperative for practitioners to select the right parameters for building robust data-driven learning models as well as use the appropriate algorithms and performance evaluation methods. In this paper, we evaluate the performance of deep learning models against the performance of other machine learning methods for wireless communication systems. We explore the different wireless communication scenarios in which deep learning can be used given Radio Frequency (RF) data, and evaluate its performance in various scenarios. Furthermore, we express it as a distribution alignment problem in which deep learning models do not perform well when learning from RF data of a particular distribution and evaluating on RF data from a different distribution. We also discuss our results in the light of how signal quality affects deep learning model leveraging on the knowledge from computer vision domain. The effect of Signal-to-Noise Ratio (SNR) selection for training on the model performance as it relates to practical implementation of deep learning in communications systems is also discussed. From our analysis, we conclude that the design and use of RF spectrum learning must be tailored to each specific scenario being considered in practice.

Nowadays, an Internet-of-Things (IoT) connected system brings a tremendous paradigm shift into the medium access control (MAC) design. In this paper, we present a distributed MAC framework assisted by machine learning for the Heterogeneous IoT system, where the IoT devices coexist with the WiFi users in the unlicensed industrial, scientific, and medical (ISM) spectrum. Specifically, the superframe is divided into two phases: a rendezvous phase and a transmission phase. During the rendezvous phase, the gateway that is capable of machine learning predicts the number of WiFi users and the IoT devices by performing a triangular handshake on the primary channel. The prediction takes advantage of the deep neural network (DNN) model which is pretrained on our universal software radio peripheral (USRP2) testbed offline. The gateway allocates the frequency channels to the WiFi and IoT systems based on the inference results. Then, the IoT devices and WiFi users initiate data transmissions during the transmission phase. Furthermore, system throughput is analyzed and optimized in two typical scenarios, respectively. An optimized MAC framework is proposed to maximize the total system throughput by finding the key design parameters. The analytical and simulation results that are conducted using the ns-2 demonstrate the effectiveness of the proposed MAC framework.

Analog hardware implemented deep learning models are promising for computation and energy constrained systems such as edge computing devices. However, the analog nature of the device and the many associated noise sources will cause changes to the value of the weights in the trained deep learning models deployed on such devices. In this study, systematic evaluation of the inference performance of trained popular deep learning models for image classification deployed on analog devices has been carried out, where additive white Gaussian noise has been added to the weights of the trained models during inference. It is observed that deeper models and models with more redundancy in design, such as VGG, are more robust to the noise in general. Also, it is observed that the performance is affected by the design philosophy of the model, the detailed structure of the model, the exact machine learning task, as well as the datasets.

The outbreak of COVID-19 has become a threat to the entire world. Wearing a face mask is an effective method to prevent from droplet spreading of COVID-19 in public areas. Face mask detection is an emerging technique that can be utilized on IoT devices to monitor and remind people to wear face mask. This paper focuses on face mask detection using deep learning on IoT devices. Specifically, four convolutional neural networks, namely, MobileNet V2, Inception V3, VGG 16, and ResNet 50 are implemented for face mask detection and tested by running inference on mobile GPU powered IoT devices including NVIDIA Jetson TX2 and NVIDIA Jetson Nano. Performance comparison and analysis are carried out using various sizes of training data, and the experimental results demonstrate that these deep learning models are robust to various size and quality of the available training data.

In this article, we propose a transfer learning (TL) enabled edge convolutional neural network (CNN) framework for 5G industrial edge networks with privacy-preserving characteristic. In particular, the edge server can use the existing image dataset to train the CNN in advance, which is further fine-tuned based on the limited datasets uploaded from the devices. With the aid of TL, the devices that are not participating in the training only need to fine-tune the trained edge-CNN model without training from scratch. Due to the energy budget of the devices and the limited communication bandwidth, a joint energy and latency problem is formulated, which is solved by decomposing the original problem into an uploading decision subproblem and a wireless bandwidth allocation subproblem. Experiments using ImageNet demonstrate that the proposed TL-enabled edge-CNN framework can achieve almost 85% prediction accuracy of the baseline by uploading only about 1% model parameters, for a compression ratio of 32 of the autoencoder.

In this paper, we propose a reconfigurable intelligent surface (RIS) assisted medium access control (MAC) protocol, termed RIS-MAC, which is used for uplink traffics and low power communications of a large number of users in Internet-of-Things (IoT) networks. In the proposed RIS-MAC protocol, RIS configuration parameters can be reserved for each user by introducing the negotiation phase before the real data transmissions. As the real data transmissions commencing, the RIS can be controlled to focus on the transceiver according to the reserved results and assist its energy-efficient data transmissions. Based on the proposed RIS-MAC, the problem is formulated to maximize the overall capacity by optimizing the negotiated parameters of RIS-MAC. The performance evaluations demonstrate the feasibility and effectiveness of RIS-MAC.

Coronavirus disease 2019 (COVID-19) is an ongoing global pandemic in over 200 countries and territories, which has resulted in a great public health concern across the international community. Analysis of X-ray imaging data can play a critical role in timely and accurate screening and fighting against COVID-19. Supervised deep learning has been successfully applied to recognize COVID-19 pathology from X-ray imaging datasets. However, it requires a substantial amount of annotated X-ray images to train models, which is often not applicable to data analysis for emerging events such as COVID-19 outbreak, especially in the early stage of the outbreak. To address this challenge, this paper proposes a two-path semi-supervised deep learning model, ssResNet, based on Residual Neural Network (ResNet) for COVID-19 image classification, where two paths refer to a supervised path and an unsupervised path, respectively. Moreover, we design a weighted supervised loss that assigns higher weight for the minority classes in the training process to resolve the data imbalance. Experimental results on a large-scale of X-ray image dataset COVIDx demonstrate that the proposed model can achieve promising performance even when trained on very few labeled training images.

International audience; Reconfigurable intelligent surface (RIS) has become a promising technology for enhancing the reliability of wireless communications, which is capable of reflecting the desired signals through appropriate phase shifts. However, the intended signals that impinge upon an RIS are often mixed with interfering signals, which are usually dynamic and unknown. In particular, the received signal-to-interference-plusnoise ratio (SINR) may be degraded by the signals reflected from the RISs that originate from nonintended users. To tackle this issue, we introduce the concept of intelligent spectrum learning (ISL), which uses an appropriately trained convolutional neural network (CNN) at the RIS controller to help the RISs infer the interfering signals directly from the incident signals. By capitalizing on the ISL, a distributed control algorithm is proposed to maximize the received SINR by dynamically configuring the active/inactive binary status of the RIS elements. Simulation results validate the performance improvement offered by deep learning and demonstrate the superiority of the proposed ISL-aided approach.

With the proliferation of Internet of Things (IoT), large amount of data are generated at edge devices with an unprecedented speed. In order to protect the privacy and security of big edge data, as well as reduce the communications cost, it is desirable to process the data locally at the edge devices. In this study, the inference performance of several popular pre-trained convolutional neural networks on three edge computing devices are evaluated. Specifically, MobileNetV1 & V2 and InceptionV3 models have been tested on NVIDIA Jetson TX2, Jetson Nano, and Google Edge TPU for image classification. Furthermore, various compression techniques including pruning, quantization, binarized neural network, and tensor decomposition are applied to reduce the model complexity. The results will provide a guidance for practitioners when deploying deep learning models on resource constrained edge devices for near real-time and on-site learning.

With the widespread use of the Internet, cybersecurity is a significant challenge faced by the world. Because of the tremendous amount of internet traffic and increased network complexity, it becomes overwhelming for network analysts to manually monitor the traffic flows and to identify intrusions in large networks. In order to adequately and effectively analyze network traffic for intrusions, multiple machine learning based intrusion detection models were proposed to detect intruders using packet traces captured in the network. It was shown that with proper training, the machine learning models could identify malicious packets accurately. In addition, data pre-processing has been performed to mitigate the problem of unbalanced datasets. Experiments show improved performance as expected. Furthermore, a multi-class classifier was built to classify not only malicious or benign traffic but also to extend labels upon the malicious data. This insures the multiclass classifiers could classify each malicious packet as a specific type of attack such as DDOS, BOTNET, and more.

Single-cell RNA sequencing (scRNAseq) enables the profiling of the transcriptomes of individual cells, thus characterizing the heterogeneity of biological samples since scRNAseq experiments are able to yield high volumes of data. Analyzing scRNAseq data will be beneficial for obtaining knowledge on cancer drug resistance, gene regulation in embryonic development, and mechanisms of stem cell differentiation and reprogramming. One common goal of scRNAseq data analytics is to identify the cell type of each individual cell that has been profiled. However, data sparsity is the main challenge due to limitations of current single-cell RNA sequencing techniques. In this paper, a novel method of representing the genes as gene embeddings is proposed to reduce data sparsity of scRNAseq data for cell type identification, which is inspired by similarities between gene system and natural language system. It contains two steps: 1) transform gene sequences into gene sentences by ranking genes in terms of their expression values; 2) employ the word2vec technique to learn gene embeddings on these gene sentences. Then we build three deep learning models, namely RNNs, Attention RNNs, and Bi-directional LSTM RNNs, for cell type classification. The proposed method is evaluated on macosko2015, a large scale scRNAseq dataset with ground truth of individual cell types. Experimental results show that the proposed method performs effectively and efficiently on identifying cell types on scRNAseq data, and it can achieve promising performance even learning on limited number of genes.

Background The use of intermittent high mechanical index (HMI) ultrasound impulses in combination with intravenously administered microbubbles (MB) has been shown to dissolve arterial and microvascular thrombi. The recent development of liquid droplets with nanoscale diameter from commercially available microbubbles may optimize thrombus permeation and subsequent thrombus dissolution. Methods Thrombi were formed from fresh porcine arterial whole blood mixed with thrombin, and placed in a vascular system mimicking branching epicardial coronary circulation (Figure 1A) at 37 Celsius and flow rate. A diagnostic ultrasound system with a tissue mimicking phantom was placed on top of the thrombus to simulate transthoracic echocardiography. Thrombus treatment consisted of a 10-minute infusion of either diluted nanodroplets (ND) or MB (same concentration) with intermittent HMI fundamental multipulse (3 usec pulse duration; FUS) or single pulse harmonic (HUS) impulses. All treatments were randomized and compared with HMI impulses alone. Efficacy was evaluated by percentage thrombus dissolution (%TD). Transmission electron microscopy (TEM) of residual thrombi after treatment was performed to examine for droplet permeation and resultant formed microbubble size. Results A relatively monodisperse nanodroplets size distribution was formed from microbubble cooled compression (mean diameter was 147±54 nm). A total of 60 porcine arterial thrombi were tested, 20 in each MB and ND treated group (20 control). Overall, there was a significant difference in %TD between ND treated thrombi (combining HUS and FUS HMI treated groups) versus MB treated thrombi and control (p Conclusion ND in combination with intermittent HMI ultrasound resulted in significantly greater thrombolysis compared to MB. In-vivo studies on nanodroplet mediated sonothrombolysis for vascular and microvascular thrombi should be performed with this modification of a commercially available microbubble. Figure 1 Funding Acknowledgement Type of funding source: Foundation. Main funding source(s): Theodore F. Hubbard Foundation

Timely and accurate screening/testing is crucial to fighting COVID-19. Compared to commonly used reverse transcriptase polymerase chain reaction (RT-PCR), chest radiography imaging (X-ray) is also a reliable, practical and rapid method to diagnose and assess COVID-19. In this paper, two types of deep learning models, namely, Convolutional Neural Networks (CNN) and Residual Neural Networks (ResNet) have been designed and tested for accurate diagnosis of COVID-19 with chest X-ray images. Experimental results demonstrate the effectiveness of the proposed approach.

Background: Resting myocardial perfusion (MP) and wall motion (WM) imaging during real-time myocardial contrast echocardiography (MCE) improves the detection of coronary artery disease (CAD). However, its prognostic role in different clinical settings (emergency department and outpatient setting) remains unclear. Methods: A systematic search in PubMed and Embase databases, and the Cochrane library, was conducted to evaluate the role of resting MP and WM in predicting major adverse cardiac events (MACE), including death, nonfatal myocardial infarction (NFMI) and urgent revascularization in patients presenting to either outpatient clinics or emergency departments with suspected symptomatic CAD. Summary receiver operating characteristic (SROC) curves, sensitivity and specificity plots were applied to assess diagnostic performance using RevMan 5.3. Results: Seven studies met criteria, including 3668 patients (six with follow up ranging from 2 days to 2.6 years). The Relative Risk (RR) for predicting MACE in patients with both abnormal resting MP and WM was 6.1 (95% CI, 5.1–7.2) and 14.3 (95% CI, 10.3–19.8) for death/NFMI, when compared to normal resting MP and WM patients. Having both abnormal resting MP and WM was also more predictive of MACE (RR, 1.7; 95% CI 1.5–1.9) and death/NFMI (RR, 2.2; 95% CI, 1.8–2.7) when compared to abnormal WM with normal resting MP. Conclusion: In this meta-analysis of both ED and outpatient clinic presentations for suspected CAD, having both a resting regional MP and WM abnormality identifies the highest risk patient for adverse events.

Background: Although small trials have detected microvascular obstruction (MVO) with variable frequency following restoration of epicardial blood flow, the independent impact of abnormal microvascular perfusion (MVP) in predicting patient outcome following emergent percutaneous coronary intervention in acute ST-segment–elevation myocardial infarction is unknown. The study aims to determine the impact of abnormal MVP following successful epicardial recanalization in ST-segment–elevation myocardial infarction. Methods: MVP was analyzed by low mechanical index ultrasound imaging within 48 hours of emergent percutaneous coronary intervention in 297 patients with acute ST-segment–elevation myocardial infarction who had restoration of Thrombolysis in Myocardial Infarction grade 3 flow in the infarct vessel. Patients were divided into normal segmental replenishment (normal MVP) after high mechanical index impulses versus delayed replenishment but normal plateau intensity (delayed MVP) and both delayed replenishment and reduced plateau intensity (MVO by definition). Demographic variables, left ventricular ejection fraction change, and 5-year follow-up of death, recurrent myocardial infarction, and congestive heart failure were analyzed. Results: MVO was seen in 115 patients (39%), delayed MVP in 124 (42%), and normal MVP in 58 patients (19%). Patients with MVO had significant lower left ventricular ejection fraction change (39±12%) at hospital discharge compared with delayed MVP (50±10%; P =0.003) and normal MVP (57±8%; P P =0.18). Both delayed MVP and MVO were independent predictors of adverse events at follow-up (hazard ratio, 21 [CI, 4–116]; P =0.001 and hazard ratio, 30 [CI, 5–183]; P Conclusions: Reduced or absent MVP following successful percutaneous coronary intervention in acute ST-segment–elevation myocardial infarction is common and associated with significantly worse outcome even with Thrombolysis in Myocardial Infarction 3 flow in the infarct vessel.

Background Electronic Medical Record (EMR) comprises patients’ medical information gathered by medical stuff for providing better health care. Named Entity Recognition (NER) is a sub-field of information extraction aimed at identifying specific entity terms such as disease, test, symptom, genes etc. NER can be a relief for healthcare providers and medical specialists to extract useful information automatically and avoid unnecessary and unrelated information in EMR. However, limited resources of available EMR pose a great challenge for mining entity terms. Therefore, a multitask bi-directional RNN model is proposed here as a potential solution of data augmentation to enhance NER performance with limited data. Methods A multitask bi-directional RNN model is proposed for extracting entity terms from Chinese EMR. The proposed model can be divided into a shared layer and a task specific layer. Firstly, vector representation of each word is obtained as a concatenation of word embedding and character embedding. Then Bi-directional RNN is used to extract context information from sentence. After that, all these layers are shared by two different task layers, namely the parts-of-speech tagging task layer and the named entity recognition task layer. These two tasks layers are trained alternatively so that the knowledge learned from named entity recognition task can be enhanced by the knowledge gained from parts-of-speech tagging task. Results The performance of our proposed model has been evaluated in terms of micro average F-score, macro average F-score and accuracy. It is observed that the proposed model outperforms the baseline model in all cases. For instance, experimental results conducted on the discharge summaries show that the micro average F-score and the macro average F-score are improved by 2.41% point and 4.16% point, respectively, and the overall accuracy is improved by 5.66% point. Conclusions In this paper, a novel multitask bi-directional RNN model is proposed for improving the performance of named entity recognition in EMR. Evaluation results using real datasets demonstrate the effectiveness of the proposed model.

Background/Aims: Cardiac fibrosis is a pathological change leading to cardiac remodeling during the progression of myocardial ischemic diseases, and its therapeutic strategy remains to be explored. S100A4, a calcium-binding protein, participates in fibrotic diseases with an unclear mechanism. This study aimed to investigate the role of S100A4 in cardiac fibrosis. Methods: Cardiac fibroblasts from neonatal C57BL/6 mouse hearts were isolated and cultured. Myocardial infarction was induced by ligating the left anterior descending coronary artery (LAD). The ligation was not performed in the sham group. A volume of 5×105pfu/g adenovirus or 5 µM/g ICG-001 was intramyocardially injected into five parts bordering the infarction zone or normal region. We used Western blotting, quantitative RT-PCR, immunofluorescence, immunohistochemistry and Masson’s trichrome staining to explore the function of S100A4. Results: We found significant increases of S100A4 level and cardiac fibrosis markers, and β-catenin signaling activation in vitro and in vivo. In addition, knockdown of S100A4 significantly reduced cardiac fibrosis and β-catenin levels. Moreover, the expression of S100A4 decreased after ICG-001 inhibited β-catenin signal pathway. Conclusion: Downregulation of S100A4 alleviates cardiac fibrosis via Wnt/β -catenin pathway in mice. S100A4 may be a therapeutic target of cardiac fibrosis.

In this paper, we propose Device Authentication Code (DAC), a novel method for authenticating IoT devices with wireless interface by exploiting their radio frequency (RF) signatures. The proposed DAC is based on RF fingerprinting, information theoretic method, feature learning, and discriminatory power of deep learning. Specifically, an autoencoder is used to automatically extract features from the RF traces, and the reconstruction error is used as the DAC and this DAC is unique to the device and the particular message of interest. Then Kolmogorov-Smirnov (K-S) test is used to match the distribution of the reconstruction error generated by the autoencoder and the received message, and the result will determine whether the device of interest belongs to an authorized user. We validate this concept on two experimentally collected RF traces from six ZigBee and five universal software defined radio peripheral (USRP) devices, respectively. The traces span a range of Signalto- Noise Ratio by varying locations and mobility of the devices and channel interference and noise to ensure robustness of the model. Experimental results demonstrate that DAC is able to prevent device impersonation by extracting salient features that are unique to any wireless device of interest and can be used to identify RF devices. Furthermore, the proposed method does not need the RF traces of the intruder during model training yet be able to identify devices not seen during training, which makes it practical.

BACKGROUND: Atrial fibrosis is involved in non-paroxysmal atrial fibrillation (NPAF) and is mainly mediated by the calcium-binding protein S100A4. This study aimed to verify the role of circulating S100A4 in the diagnosis of atrial fibrosis and the prognosis of NPAF. METHODS: Consecutive NPAF patients undergoing catheter ablation were selected. Patients with low voltage amplitudes (

This paper presents an optimized equivalent consumption minimization strategy (ECMS) for four-wheel-drive (4WD) hybrid electric vehicles (HEVs) incorporating vehicle connectivity. In order to be applicable to the 4WD architecture, the ECMS is designed based on a rule-based strategy and used under the condition that a certain propulsion mode is activated. Assuming that a group of 4WD HEVs are connected and position information can be shared with each other, we formulate a decentralized model predictive control (MPC) framework that compromises fuel efficiency, mobility, and inter-vehicle distance to optimize the velocity profile of each individual vehicle. Based on the optimized velocity profile, an optimization problem considering both fuel economy and battery state of charge (SOC) sustainability is formulated to optimize the equivalent factors (EFs) of the ECMS for HEVs over an appropriate time window. MATLAB User Datagram Protocol (UDP) is used in the codes run on multiple computers to simulate the wireless communication among vehicles, which share position information via UDP-based communication, and dSPACE is used as a software-in-the-loop platform for the simulation of the optimized ECMS. Simulation results validate the control effectiveness of the proposed method.

This paper presents a decentralized fuel efficient model predictive control (MPC) strategy for a group of connected vehicles incorporating vertical vibration. To capture the vehicle vibration dynamics, the dynamics of the suspension system is integrated with the longitudinal dynamics of the vehicle. Furthermore, a MPC framework with finite time horizon is formulated to calculate the optimal velocity profile that compromises fuel economy, mobility and ride comfort for every individual vehicle with the safety and physical constraints considered. In the MPC framework, the target velocity is calculated using signal phase and timing (SPAT) information to reduce the number of stoppage at red lights, and the vertical acceleration is calculated parallel to the calculation of the fuel consumption. The MPC optimal problem is solved with fast-MPC approach which enhances the computational efficiency via exploiting the structure of the control system and approximate methods. Simulation studies are conducted over different SPATs and connectivity penetration rates and the results validate the advantages of the proposed control architecture.

Multi-cell thin-walled tubes have proven to be better in energy absorption than plain square tubes subjected to axial compression. Therefore, square multi-cell structures have been extensively utilized as energy absorbers in automobiles. This paper provides an investigation on the crashworthiness of rectangular single-, double- and triple-cell columns under axial loading and an optimization design of rectangular unequal triple-cell tubes. First, a theoretical solution is derived for the mean crushing force (MCF) of tubes with unequal triple-cell configuration. Second, quasi-static crushing experiments and finite element analyses (FEA) are conducted on single-, double- and unequal triple-cell columns. Theoretical predictions compare well with experimental and numerical data, and all results show that the triple-cell tubes exhibit the best crashworthiness among all the samples. Third, in order to study effects of wall thickness distribution and the layout of internal ribs on crashing behavior, multiobjective optimization design is implemented combining Radial Basis Function (RBF) model with Non-dominant sorting Genetic Algorithm II (NSGA-II). The optimal solution obtained from Pareto frontier indicates that unequal triple-cell tube with appropriate thickness distribution and arrangement of internal ribs is superior in energy absorption to initial design.

This study presents a novel decentralised hierarchical global energy management control strategy for a group of connected four-wheel-drive hybrid electric vehicles (HEVs) in urban road conditions. In the higher level controller, signal phase and timing information and the optimal cruising velocity are combined to generate the target velocities for the HEVs. A model predictive control framework that focuses on the tracking of the target velocity and the associated desired control variable for every individual vehicle is proposed for the prediction of the optimal velocity that compromises fuel economy, mobility and safety. In the lower level controller, a dynamic programming problem is formulated that utilises the predicted velocity for the global energy management optimisation of every individual HEV. Simulation results validate the advantages of the proposed higher and lower level controllers.

A hierarchical energy management control strategy was proposed to realize the global optimal energy management for a group of four-wheel-drive hybrid electric vehicles (4WD HEVs) in connected vehicle environment. The higher level controller utilized signal phase and timing (SPAT) for the evaluation of the target velocities and optimal target velocity prediction objective function model was established based on model predictive control (MPC). The lower level controller adopted dynamic programming (DP) for the energy management of the vehicles. dSPACE-based simulation results validate that the higher level controller can avoid the 4WD HEVs from red light stopping and avoid collisions of the vehicles. Compared with the Gipps car following model, the fuel economy is increased by 21.4%, 21.5% and 20.6% when the lower level controller is based on DP, ECMS and the rule-based strategy, respectively. In addition, the DP-based lower level controller can achieve good velocity tracking and SOC balance. Compared with ECMS and rule-based control strategies, the average fuel economy for the group of 4WD HEVs with DP-based strategy is improved by 7.8% and 15.9% , respectively.

Wireless big data describes a wide range of massive data that is generated, collected and stored in wireless networks by wireless devices and users. While these data share some common properties with traditional big data, they have their own unique characteristics and provide numerous advantages for academic research and practical applications. This article reviews the recent advances and trends in the field of wireless big data. Due to space constraints, this survey is not intended to cover all aspects in this field, but to focus on the data aided transmission, data driven network optimization and novel applications. It is expected that the survey will help the readers to understand this exciting and emerging research field better. Moreover, open issues and promising future directions are also identified.

Social media is a popular platform for timely information sharing. One of the important challenges for social media platforms like Twitter is whether to trust news shared on them when there is no systematic news verification process. On the other hand, timely detection of rumors is a non-trivial task, given the fast-paced social media environment. In this work, we proposed an ensemble model, which performs majority-voting on a collection of predictions by deep neural networks using time-series vector representation of Twitter data for timely detection of rumors. By combining the proposed data pre-processing method with the ensemble model, better performance of rumor detection has been demonstrated in the experiments using PHEME dataset. Experimental results show that the classification performance has been improved by 7.9% in terms of micro F1 score compared to the baselines., Comment: 12 pages, 3 figures

In this paper, we propose a distributed deep reinforcement learning (DRL) based medium access control (MAC) protocol, termed DRL-MAC, which is used to assist the backscatter communications for Internet-of-Things (IoT) networks. By leveraging the current Wi-Fi infrastructure, the backscatter communications can be reserved in advance to avoid the interference from Wi-Fi communications. In the proposed MAC protocol, the deep reinforcement learning is further introduced to learn the reserved information and make decisions such as 1) which backscatter device (TAG) will be serviced, and 2) the reservation step for the serviced TAG. In addition, the utility function is defined and the optimization problem is formulated to balance the backscatter communications and Wi-Fi communications. Moreover, a DRL algorithm is proposed to obtain the optimal strategy. The numerical results show the effectiveness of the proposed MAC for backscatter communications.

In future wireless systems, intelligent capabilities are of utmost importance. To efficiently utilize resources, communication systems require knowledge of the prevalent situation in a frequency band through learning. To learn appropriately, it is critical for practitioners to select the right parameters in building learning models, use the appropriate algorithms and performance evaluation methods. In this paper, we evaluate the performance of some deep learning models compared to other machine learning methods, explore the different scenarios in which deep learning can be used for radio frequency (RF) monitoring, and evaluate performance in the various scenarios. Our work looks at the best practices and procedures for developing intelligent RF Learning. Specifically, we analysed over-the-air RF dataset collected from a USRP-based testbed to identify the number of interfering devices as a case study. From the obtained results, we discuss how Signal-to-Noise Ratio (SNR) selection for training affects the model performance as it relates to practical implementation of Deep Learning in communications systems.

Localization of aircraft is important to control air traffic safely and effectively. Although the Automatic Dependent Surveillance Broadcast (ADS-B) has many advantages, the transfer of control over the reported location to the aircraft brings a number of safety and security issues. In order to mitigate these issues and determine the locations of the aircraft which do not have position reporting capabilities or may report wrong locations, complementary or redundant localization methods that are independent of the aircraft are needed. The goal of this paper is to study the feasibility to localize aircraft (estimate the longitude, latitude, and altitude of an aircraft) based on crowdsourced air traffic control communication data, specifically time of arrival and signal strength measurements reported by many different sensors. Specifically, we design and test a deep neural network model for aircraft location prediction using realworld data from OpenSky Network, a crowd-sourced receiver network that obtains volumes of air traffic data from thousands of sensors. It is demonstrated that the proposed deep neural network outperforms the time difference of arrival (TDOA) and support vector regressor (SVR) in terms of the mean absolute percentage error (MAPE), and the proposed deep learning based method using crowdsourced air traffic control communication data is an effective solution for accurate aircraft location prediction that are independent of the aircraft.

False information has become a weapon in cyber-warfare. How to detect false information effectively and efficiently on social media is a challenging problem. In this study, a novel method of rumor detection on Twitter tweets is proposed as a proof-of-concept for fast detection of false information on social media. Specifically, the proposed method will use the propagation pattern of the tweets to detect false information rather than the contents. As a result, the proposed method is very effective in reducing the dimensionality of the input feature set, and it requires much less computational time compared to content-based methods. Extensive experiments on PHEME dataset, a collection of Twitter rumors and non-rumors posted during five breaking news, have been performed to demonstrate the effectiveness of the proposed method. We also observe that deep learning models such as recurrent neural networks outperform classical machine learning models in terms of micro-F score.

Aims To evaluate the prognostic value of myocardial perfusion (MP) imaging during contrast stress echocardiography (cSE) in patients with known or suspected coronary artery disease (CAD). Methods and results A search in PubMed, Embase databases, and the Cochrane library was conducted through May 2019. The Cochran Q statistic and the I2 statistic were used to assess heterogeneity, and the results were analysed by RevMan V5.3 and Stata V15.1 software. Twelve studies (seven dipyridamole and five exercise/dobutamine) without evidence of patient overlap (same institution publishing results over a similar time period) enrolling 5953 subjects (47% female, 8–80 months of follow-up) were included in the analysis. In all studies, total adverse cardiovascular events were defined as either cardiac death, non-fatal myocardial infarction (NFMI), or need for urgent revascularization. Hazard ratios (HRs) revealed that a MP abnormality [pooled HR 4.75; 95% confidence interval (CI) 2.47–9.14] was a higher independent predictor of total events than abnormal wall motion (WM, pooled HR 2.39; 95% CI 1.58–3.61) and resting left ventricular ejection fraction (LVEF, pooled HR 1.92; 95% CI 1.44–2.55) with significant subgroup differences (P = 0.002 compared with abnormal WM and 0.01 compared with abnormal LVEF). Abnormal MP was associated with higher risks for death [Risk ratio (RR) 5.24; 95% CI 2.91–9.43], NFMI (RR 3.09; 95% CI 1.84–5.21), and need for coronary revascularization (RR 16.44; 95% CI 6.14–43.99). Conclusion MP analysis during stress echocardiography is an effective prognostic tool in patients with known or suspected CAD and provides incremental value over LVEF and WM in predicting clinical outcomes.

Smart metering has grabbed significant attention in recent years, particularly for the utility providers who plan the energy resources and take control actions to balance the electricity demand and supply by load forecasting. Currently, load forecasting is performed at the aggregated level, not at an individual level because it is highly uncertain and complex. Specifically, the performance of short-term forecasting is affected significantly by the variance of load uncertainty. Moreover, limited work has been done to help users choose the optimal usage plan. In this paper, we evaluate several deep learning models for load forecasting. In addition, we employ deep learning techniques to provide the optimal power plan for users based on their power usage. Experimental results using the data from the Irish Social Science Data Archive demonstrate the effectiveness of the proposed schemes.

Internet of Things (IoT) and 4G/5G wireless networks have added huge number of devices and new services, where commercial-of-the-shelf (COTS) IoT devices have been deployed extensively. To ensure secure operations of these systems with wireless transmission capabilities, Radio Frequency (RF) surveillance is important to monitor their activities in RF spectrum and detect unauthorized IoT devices. Specifically, in order to prevent an adversary from impersonating legitimate users using identical devices from the same manufacturer, unique “signatures” must be obtained for every individual device in order to uniquely identify each device. In this study, a novel intrusion detection method is proposed to detect unauthorized IoT devices using deep learning. The proposed method is based on RF fingerprinting since physical layer based features are device specific and more difficult to impersonate. RF traces are collected from six “identical” ZigBee devices via a USRP based test bed. The traces span a range of Signal-to-Noise Ratio, to ensure robustness of the model. A convolutional neural network is used to extract features from the RF traces, and dimension reduction and de-correlation are performed on the extracted features. The reduced features are then clustered to identify IoT devices. We show that the proposed method is able to identify devices that are not observed during training. The results not only highlight the benefit of deep learning based feature extraction, but also show promising prospects for being able to distinguish new devices (classes) that are not observed during training.