Search

Today, cloud computing technology has attracted the attention of many researchers. According to the needs of users to quickly execute requests and provide quality services, optimal allocation of resources and timing of task execution between virtual machines in cloud computing are of great importance. One of the important challenges that cloud service providers face is the effective management of resources by physical infrastructure. Therefore, in this paper, an autonomous system based on the Clipped Double Deep Q-Learning (CDDQL) Algorithm and the meta-heuristic Particle Swarm Optimization (PSO) for resource allocation is proposed in the Fog-cloud computing infrastructure. The PSO algorithm is used to prioritize the tasks and CDDQL is used as the core of the autonomous system (Auto-CDDQL) to allocate the desired VM resources to the tasks. The proposed Auto-CDDQL is implemented in the Fog and performs this process autonomously. By evaluating the results, it was observed that the amount of Make Span, response time, task completion, resource utilization, and energy consumption rate in the proposed AutoCDDQL on the c-hilo dataset, compared to the FCFS, RR, and PBTS methods, are significantly improved. [ABSTRACT FROM AUTHOR]

Machine learning techniques aim to mimic the human ability to automatically learn how to perform tasks through training examples. They have proven capable of tasks such as prediction, learning and adaptation based on experience and can be used in virtually any scientific application, ranging from biomedical, robotic, to business decision applications, and others. However, the lack of domain knowledge for a particular application can make feature extraction ineffective or even unattainable. Furthermore, even in the presence of pre-processed datasets, the iterative process of optimizing Machine Learning parameters, which do not translate from one domain to another, maybe difficult for inexperienced practitioners. To address these issues, we present in this paper a Vectorized Automated ML Pre-processIng and post-pRocEssing framework, approximately named (VAMPIRE), which implements feature extraction algorithms capable of converting large time-series recordings into datasets. Also, it introduces a new concept, the Activation Engine, which is attached to the output of a Multi Layer Perceptron and extracts the optimal threshold to apply binary classification. Moreover, a tree-based algorithm is used to achieve multi-class classification using the Activation Engine. Furthermore, the internet of things gives rise to new applications such as remote sensing and communications, so consequently applying Machine Learning to improve operation accuracy, latency, and reliability is beneficial in such systems. Therefore, all classifications in this paper were performed on the edge in order to reach high accuracy with limited resources. Moreover, forecasts were applied on three unrelated biomedical datasets, and on two other pre-processed urban and activity detection datasets. Features were extracted when required, and training and testing were performed on the Raspberry Pi remotely, where high accuracy and inference speed were achieved in every experiment. Additionally, the board remained competitive in terms of power consumption when compared with a laptop which was optimized using a Graphical Processing Unit. [ABSTRACT FROM AUTHOR]

Session-based recommendation aims to predict the next item a user may interact with based on sessions' information. Most existing session-based recommendation models only rely on the information of item-to-item transitions. However, we argue that there exists another type of information in a session that can be mined and made use of, i.e., the information of item-to-item relevance, which could help increase a session-based recommendation model's performance. In this case, a model named Relevance-Aware Graph Neural Network (RA-GNN) is proposed in this paper, which captures and utilizes both types of information carried in a session, i.e., both the information of item-to-item transitions and the information of item-to-item relevance. Specifically, our paper first proposes to construct a relevance graph and transition graphs, respectively, to capture two types of information. Then two different embedding modules are designed to learn transition and relevance embeddings for items, respectively. Regarding the embeddings for sessions, to obtain the long-term and short-term interest of users, we propose a module to derive the global and local embeddings of sessions, which are further combined to obtain the session embeddings. Experiments are conducted on three real-world datasets to compare the proposed RA-GNN with the state-of-the-art session-based recommendation models. According to the results of experiments, our proposed RA-GNN outperforms the comparison models. The results of experiments show that the proposed model could effectively capture both types of sessions' information and obtain the interest of users more accurately, therefore improving the performance of session-based recommendation. [ABSTRACT FROM AUTHOR]

Big data frameworks such as Storm, Spark and Hadoop are widely deployed in commercial and research applications, the energy consumption of cloud data centers that support big data processing platforms is becoming more and more prominent. However, job scheduling is a complex problem in the presence of various service level agreement (SLA) goals, such as cost reduction and job performance improvement. The highly heterogeneous nature of clusters and the variability of resource requirements acrossworkloads make energy-efficient scheduling on big data platforms extremely complex under SLA constraints. Existing performance-based models and heuristic scheduling methods rely excessively on historical data and are difficult to optimize or modify for changes in load and clusters. In this paper, we construct an energy consumption model based on resource utilization and a reinforcement learning model for energy-efficient scheduling under SLA constraints for Spark clusters, and design two Deep Reinforcement Learning (DRL) algorithms. The cluster scheduler designed and implemented based on this model can automatically capture different load characteristics and inherent cluster characteristics, find the appropriate executor creation policy for resource allocation, and reduce cluster energy consumption under the constraint of job execution time. Experimental results show that the DRL scheduler proposed in this paper saves a maximum energy of about 33% under different load characteristics. [ABSTRACT FROM AUTHOR]

Path finding is used to solve the problem of finding a traversable path through an environment with obstacles. This problem can be seen in many different fields of study and these areas rely on fast and efficient path finding algorithms. This paper aims to describe and review state of the art optimization techniques that are used on optimized path finding and compare their performances. Moreover, a special attention is paid on the proposed approaches to identify how they are tested on different test cases; whether the test cases are automatically generated or benchmark instances. The review opens avenues about the importance of automatic test case generation to test the different path finding algorithms. [ABSTRACT FROM AUTHOR]

Processing of data gathered from new communication devices, such as Internet of Things (IoT)-based technology, has grown dramatically in the past decade. Resource management plays a vital role in cloud/fog-based platforms' efficiency. Alternatively, a deadline-based workflow scheduling mechanism is an approach to resource management that increases cloud/fog computing efficiency. However, most proposed methods may overload some resources and underload others. Consequently, adopting a proper load-balancing approach has a major impact on optimizing Quality of Service (QoS) and improving customer satisfaction. This paper presents a 4-layer software architecture for analyzing workflows and dynamic resources in cloud/fog/IoT environments to address such a problem. This approach also considers workload and presence prediction of IoT nodes as dynamic resources. Moreover, the 4 + 1 architectural view models represent architecture layers, components, and significant interactions. Architecture components are ultimately proposed to meet quality attributes such as availability, reliability, performance, and scalability. The proposed architecture evaluation is according to the Architecture Tradeoff Analysis Method (ATAM) as a scenario-based technique. Compared with previous works, various scenarios, and more quality attributes are discussed within this evaluation, in addition to analyzing and predicting workload and the presence prediction of dynamic resources. [ABSTRACT FROM AUTHOR]

Statistical partitioning and publishing is a common strategy employed in location-based big data services. This paper proposes a deep learning-based statistical partitioning structure prediction method and a differential publishing method to address the problems of unreasonable structures and low efficiency in traditional statistical partitioning and publishing of location-based big data. The 2D space has been finely partitioned and merged bottom-up to construct a reasonable partition structure. The partition structure matrices have been organized into a 3D spatio-temporal sequence and the spatio-temporal characteristics are extracted via a deep learning model to realize the prediction of the partition structure. Differential privacy budget allocation and Laplace noise incorporation have been implemented on the predicted partition structure to realize the privacy protection of the statistical partitioning and publishing of location-based big data. Experimental comparison of the real-world datasets suggests the advantages of the envisaged method in improving the querying accuracy of the published data and the execution efficiency of the publishing algorithm. [ABSTRACT FROM AUTHOR]

With the development of Internet technology, cyberspace security has become a research hotspot. Network traffic classification is closely related to cyberspace security. In this paper, the problem of classification based on raw traffic data is investigated. This involves the granularity analysis of packets, separating packet headers from payloads, complementing and aligning packet headers, and converting them into structured data, including three representation types: bit, byte, and segmented protocol fields. Based on this, we propose the Rew-LSTM classification model for experiments on publicly available datasets of encrypted traffic, and the results show that excellent results can be obtained when using only the data in packet headers for multiple classification, especially when the data is represented using bit, which outperforms state-of-the-art methods. In addition, we propose a global normalization method, and experimental results show that it outperforms feature-specific normalization methods for both Tor traffic and regular encrypted traffic. [ABSTRACT FROM AUTHOR]

The rapid development and deployment of network services has brought a series of challenges to researchers. On the one hand, the needs of Internet end users/applications reflect the characteristics of travel alienation, and they pursue different perspectives of service quality. On the other hand, with the explosive growth of information in the era of big data, a lot of private information is stored in the network. End users/applications naturally start to pay attention to network security. In order to solve the requirements of differentiated quality of service (QoS) and security, this paper proposes a virtual network embedding (VNE) algorithm based on deep reinforcement learning (DRL), aiming at the CPU, bandwidth, delay and security attributes of substrate network. DRL agent is trained in the network environment constructed by the above attributes. The purpose is to deduce the mapping probability of each substrate node and map the virtual node according to this probability. Finally, the breadth first strategy (BFS) is used to map the virtual links. In the experimental stage, the algorithm based on DRL is compared with other representative algorithms in three aspects: long term average revenue, long term revenue consumption ratio and acceptance rate. The results show that the algorithm proposed in this paper has achieved good experimental results, which proves that the algorithm can be effectively applied to solve the end user/application differentiated QoS and security requirements. [ABSTRACT FROM AUTHOR]

Cross-language entity alignment forms an important component of building a Knowledge Graph. The task of cross-lingual entity alignment is to match entities in a source language with their counterparts in target languages. In practice, there is an imbalance of attribute information in corresponding entities at the same level, and the problem of neighboring point weight assignment is not considered, which not only loses the association information between entities but also limits the utilization of entity attributes in the alignment process, making this task challenging. In this paper, we propose a cross-lingual entity alignment method based on Graph convolutinal neural network and Graph attention network. Specifically, it can capture more spatial information by assigning respective weights to the neighbors of different nodes through multi-level learning of entity structure, attributes, and attention. In addition, the weights of neighboring node features depend entirely on the node features, which gets rid of the dependence on the graph. The experiments show that our models outperform state-of-the-art methods at a fraction of the cost. [ABSTRACT FROM AUTHOR]

Rumor detection is a hot research topic in social networks. It is challenging to simultaneously extract content features and structural features from rumor conversations to detect rumors, and many existing methods only focus on one of them. In this paper, we propose a Source-Replies conversation Tree Convolutional Neural Net (TCN) to extract these two features simultaneously for the rumor detection task. Specifically, we first build Source-Replies conversation Trees (SR-Trees) based on rumor conversations, and then we construct an SR-Tree-based Auto-Encoder (TAE) on SR-Trees. A TAE designs Spatial Tree Convolution and Tree-pooling to build an effective feature extractor to extract content features and structural features from SR-Trees. Based on the pre-trained feature extractor in the TAE, an end-to-end TCN is proposed to detect rumors. In experiments, we verify that compared with other commonly used Rumor analysis models, the proposed TCN is effective on the rumor detection task. [ABSTRACT FROM AUTHOR]

With the rapid development in artificial intelligence and mobile networks, the past decade has witnessed the flourish of social media, and information diffusion popularity prediction in social media has attracted wide attention in both academics and industrials. However, existing popularity prediction methods either rely heavily on human experience to handcraft the features, designate the generative model, or largely depend on the underlying user relation network for embedding learning. Motivated by the above observation, this paper studies the precise prediction of the information diffusion popularity only based on early repost information, given that the underlying user relation network is unknown. To solve this problem, we propose RNe2Vec (repost network to vector), a repost network embedding-based diffusion popularity prediction algorithm. Specifically, we first build a repost network from the early repost data, and then use biased random walks to generate node sequences, in which we elaborately design walking rules to capture different repost behaviors. After that we employ the skip-gram method to learn low dimensional node vectors from the node sequences. Finally, we apply PCA (principal component analysis) algorithm on the node vectors for dimensionality reduction, and combine the embedding features with handcrafted features to train the downstream machine learning models. Experimental results on a microblog dataset show that incorporating network embedding features can significantly improve the overall prediction accuracy. [ABSTRACT FROM AUTHOR]

Vehicular Ad-hoc Networks (VANETs) are an evolving technology in the transportation industry. Reliability is a critical issues in vehicular communications to ensure road safety. An analytical framework is presented in this paper to estimate the reliability of Vehicle-to-Vehicle (V2V) communications in Cognitive Radio (CR)-VANETs. The proposed framework considers nodes' reliability and different challenges of physical, MAC, and network layers for communications' reliability, including the probability of channel availability for CR-enabled vehicles, channel fading, contention among the transmitting vehicles, hidden terminal problem, and transmission redundancy. The channel availability for reliability assessments of CR vehicular communications has not been considered in the previous research. Therefore, a Markov model is proposed to estimate the probability of channel availability for CR-enabled vehicular nodes. Also, a dataset is created using the proposed analytical framework, and the reliability estimation of the V2V communications is automated using an Artificial Neural Network (ANN) model. The analytical findings are validated by NS-3 simulations and compared with existing methods, showing the accuracy of the proposed method. Moreover, the correlation coefficient (R) and Root Mean Square Error (RMSE) of the test data for the best case of the ANN model, with one hidden layer, are 0.9948 and 0.0289, respectively. These values indicate the accurate estimation of CR-vehicular communications' reliability by the analytical framework. [ABSTRACT FROM AUTHOR]

When COVID-19 was declared as a pandemic by the World Health Organization on 11 March 2020, national governments and health authorities across the world begun considering different preventive measures to fight against the coronavirus outbreak. Researchers and tech companies worldwide have been striving to utilize advanced technologies to aid in the fight against the Covid-19 outbreak. This paper aims to couple multifunction drone with AI to deliver wireless services that will help the fight against the Coronavirus pandemic. The proposed drone-eye-system with its thermal imaging cameras and an AI framework utilizes a Convolutional Neural Network (CNN) with its Modified Artificial Neural Network (MANN) for face mask detection of people wearing masks in public. The system can perform basic diagnostic functions such as elevated body temperatures for helping minimize the risk of spreading the infection through close contact. The AI framework evolve an optimized elevation angle θ and altitude h t to enhance wireless connectivity between a drone and a ground station, which in turn leads to better throughput and power consumption. The proposed framework has been developed using the MATLAB toolbox and shows promising results with an accuracy of face mask detection of 82.63%, with an F1-score of 0.98, and an enhanced by 10% link budget parameters. [ABSTRACT FROM AUTHOR]

The increasing growth of online social networks has drawn researchers' attention to link prediction and has been adopted in many fields, including computer sciences, information science, and anthropology. The link prediction in attributed networks is a new challenge in this field, one of the interesting topics in recent years. Nodes are also accompanied in many real-world systems by various attributes or features, known as attributed networks. One of the newest methods of link prediction is embedding methods to generate the feature vector of each node of the graph and find unknown connections. The DeepWalk algorithm is one of the most popular graph embedding methods that capture the network structure using pure random walking. The present paper seeks to present a modified version of deep walk based on pure random walking for solving link prediction in the attributed network, which will be used for both network structure and node attributes, and the new random walk model for link prediction will be introduced by integrating network structure and node attributes, based on the assumption that two nodes on the network will be linked since they are nearby in the network, or connected for the reason of similar attributes. The results indicate that two nodes are more probable to establish a link in the case of possessing more structure and attribute similarity. In order to justify the proposal, the authors carry out many experiments on six real-world attributed networks for comparison with the state-of-the-art network embedding methods. The experimental results from the graphs indicate that our proposed approach is more capable compared to other link prediction approaches and increases the accuracy of prediction. [ABSTRACT FROM AUTHOR]

Although mobile technologies keep evolving through years, Fault management and cyber-security management in mobile networks are still treated as separated notions with different blocks and different approaches whereas in practice, they are highly correlated. In this paper, we propose a framework that takes into account the correlation between these two management systems. The framework is based on several prediction agents where each agent is composed of a security predictor, a fault predictor and a generic anomaly detection model. A re-enforcement process allows to enhance the reliability of the machine learning training and prediction phases of the different predictors. Besides, each agent can collaborate with its neighborhood for a more resilient network. An application of this framework to 5G architecture is proposed by mapping the components of our framework with network slices. Finally, an experimentation is held over a testbed that we set up on openstack in order to forecast future anomalies related to proxy overload, latency violation in call session network functions and to excessive usage of memory. The training is achieved with ARIMA and deep learning models with promising results. [ABSTRACT FROM AUTHOR]

With air quality issues becoming ever greater global concerns, many countries are facing numerous air pollution problems. Among all the particulate matter of air pollution, PM2.5 (whose aerodynamic diameter is 2.5 μm or less) is of particular concern. Long-term exposure to high concentrations of PM2.5 can negatively affect human health. The negative effects of PM2.5 make forecast its concentration an urgent need. In this paper, we propose a novel Factory-aware Attentional LSTM Model (FAA-LSTM) for PM2.5 air pollution predictions. The proposed model collects air pollution data from both the monitor stations and micro air quality sensors, in which a spatial transformation is designed to obtain the local area data of PM2.5 grids. Next, a novel factory-aware attention mechanism over a long short-term memory (LSTM) neural network is proposed to extract the hidden features of industrial factors and derive their factory attention weights on the influences of PM2.5 concentrations. The influence of neighboring factory data over local PM2.5 grids can be weighted to discover the importance of PM2.5 concentrations of neighboring areas. Moreover, the model combines these heterogeneous data and the global station data to forecast PM2.5 concentrations. The experiment's evaluation is conducted using both air pollution data and industrial data. The results show that the factory-aware attention mechanism helps to improve the prediction performance by exploring the effect of the factory distribution on PM2.5 pollutants in the local areas. While monitoring stations have been established to collect air quality information and forecast air quality, few studies have taken the different monitoring areas and industrial features into account. The proposed novel model considering PM2.5 concentrations from local neighboring areas and global station, and industrial data as features can effectively indicate the impact of PM2.5 pollution with industrial emissions and spatial relationships. Our research work can improve the prediction accuracy of PM2.5 and contribute to increasing the practical value for air quality management. [ABSTRACT FROM AUTHOR]

Personality describes an individual's internal attributes, encompassing distinctive thoughts, emotions, and behaviour. Various studies have been undertaken to quantify these personality traits using externally visible cues and measures. Despite the possibility of a disparity between the apparent and real personality traits, researchers have identified various deep learning techniques to measure the apparent personality as part of the ChaLearn Looking at People, ECCV challenge. We provide an overview of the different deep learning models used for apparent personality detection. The primary objective of previous studies was to develop highly accurate prediction models, and subsequently, the focus shifted towards interpreting the output of these techniques. Although explainable AI (XAI) techniques have shown that facial features are essential in predicting personality, the consistency of these results is still a critical factor to consider. Moreover, the majority of studies utilized the ECCV challenge CVPR'17 dataset to predict apparent personality; therefore, there is a requirement to validate these findings through additional datasets. In conclusion, ample scope exists for further research in the domain of personality detection that can yield highly accurate and more interpretable models. [ABSTRACT FROM AUTHOR]

With the rapid development of artificial intelligence, privacy threats are already getting the spotlight. One of the most common privacy threats is the membership inference attack (MIA). Existing MIAs can effectively explore the potential privacy leakage risks of deep neural networks. However, DNNs are usually compressed for practical use, especially for edge computing, MIA will fail due to changes in DNNs' structure or parameters during the compression. To address this problem, we propose CM-MIA, an MIA against compression models, which can effectively determine their privacy leakage risks before deployment. In specific, firstly we use a variety of compression methods to help build shadow models for different target models. Then, we use these shadow models to construct sample features and identify abnormal samples by calculating the distance between each sample feature. Finally, based on the hypothesis test, we determine whether the abnormal sample is a member of the training dataset. Meanwhile, only abnormal samples are used for membership inference, which reduces time costs and improves attack efficiency. Extensive experiments are conducted on 6 datasets to evaluate CM-MIA's attack capacity. The results show that CM-MIA achieves the state-of-the-art attack performance in most cases. Compared with baselines, the attack success rate of CM-MIA is increased by 10.5% on average. [ABSTRACT FROM AUTHOR]

In the high-performance computing (HPC) domain, federated learning has gained immense popularity. Especially in emotional and physical health analytics and experimental facilities. Federated learning is one of the most promising distributed machine learning frameworks because it supports data privacy and security by not sharing the clients' data but instead sharing their local models. In federated learning, many clients explicitly train their machine learning/deep learning models (local training) before aggregating them as a global model at the global server. However, the FL framework is difficult to build and deploy across multiple distributed clients due to its heterogeneous nature. We developed Docker-enabled federated learning (DFL) by utilizing client-agnostic technologies like Docker containers to simplify the deployment of FL frameworks for data stream processing on the heterogeneous client. In the DFL, the clients and global servers are written using TensorFlow and lightweight message queuing telemetry transport protocol to communicate between clients and global servers in the IoT environment. Furthermore, the DFL's effectiveness, efficiency, and scalability are evaluated in the test case scenario where real-time emotion state classification is done from distributed multi-modal physiological data streams under various practical configurations. [ABSTRACT FROM AUTHOR]

Incorporating machine learning into automatic performance analysis and tuning tools is a promising path to tackle the increasing heterogeneity of current HPC applications. However, this introduces the need for generating balanced datasets of parallel applications' executions and for dealing with natural imbalances for optimizing performance parameters. This work proposes a holistic approach that integrates a methodology for building balanced datasets of OpenMP code-region patterns and a way to use such datasets for tuning performance parameters. The methodology uses hardware performance counters to characterize the execution of a given region and correlation analysis to determine whether it covers an unique part of the pattern input space. Nevertheless, a balanced dataset of region patterns may become naturally imbalanced when used for training a model for tuning any specific performance parameter. For this reason, we have explored several methods for dealing with naturally imbalanced datasets for finding the appropriated way of using them for tuning purposes. Experimentation shows that the proposed methodology can be used to build balanced datasets and that such datasets, plus a combination of Random Forest and binary classification, can be used to train a model able to accurately tune the number of threads of OpenMP parallel regions. [ABSTRACT FROM AUTHOR]

In this work, we build a general piece-wise model to analyze data-parallel (DP) training costs of convolutional neural networks (CNNs) on clusters of GPUs. This general model is based on i) multi-layer perceptrons (MLPs) in charge of modeling the NVIDIA cuDNN/cuBLAS library kernels involved in the training of some of the state-of-the-art CNNs; and ii) an analytical model in charge of modeling the NVIDIA NCCL Allreduce collective primitive using the Ring algorithm. The CNN training scalability study performed using this model in combination with the Roofline technique on varying batch sizes, node (floating-point) arithmetic performance, node memory bandwidth, network link bandwidth, and cluster dimension unveil some crucial bottlenecks at both GPU and cluster level. To provide evidence of this analysis, we validate the accuracy of the proposed model against a Python library for distributed deep learning training. [ABSTRACT FROM AUTHOR]

The ongoing COVID-19 (novel coronavirus disease 2019) pandemic has triggered a global emergency, resulting in significant casualties and a negative effect on socioeconomic and healthcare systems around the world. Hence, automatic and fast screening of COVID-19 infections has become an urgent need of this pandemic. Real-time reverse transcription polymerase chain reaction (RT-PCR), a commonly used primary clinical method, is expensive and time-consuming for skilled health professionals. With the aid of various AI functionalities and advanced technologies, chest CT scans may thus be a viable alternative for quick and automatic screening of COVID-19. At the moment, significant advances in 5G cellular and internet of things (IoT) technology are finding use in various applications in the healthcare sector. This study presents an IoT-enabled deep learning-based stacking model to analyze chest CT scans for effective diagnosis of COVID-19 encounters. At first, patient data will be obtained using IoT devices and sent to a cloud server during the data procurement stage. Then we use different fine-tuned CNN sub-models, which are stacked together using a meta-learner to detect COVID-19 infection from input CT scans. The proposed model is evaluated using an open access dataset containing both COVID-19 infected and non-COVID CT images. Evaluation results show the efficacy of the proposed stacked model containing fine-tuned CNNs and a meta-learner in detecting coronavirus infections using CT scans. [ABSTRACT FROM AUTHOR]

Automatic detection of weapons is significant for improving security and well being of individuals, nonetheless, it is a difficult task due to large variety of size, shape and appearance of weapons. View point variations and occlusion also are reasons which makes this task more difficult. Further, the current object detection algorithms process rectangular areas, however a slender and long rifle may really cover just a little portion of area and the rest may contain unessential details. To overcome these problem, we propose a CNN architecture for Orientation Aware Weapons Detection, which provides oriented bounding box with improved weapons detection performance. The proposed model provides orientation not only using angle as classification problem by dividing angle into eight classes but also angle as regression problem. For training our model for weapon detection a new dataset comprising of total 7801 weapons images is gathered from the web and then manually annotated with position oriented bounding boxes. Our dataset provides not only oriented bounding box as ground truth but also horizontal bounding box. We also provide our dataset in multiple formats of modern object detectors for further research in this area. The proposed model is evaluated on this dataset, and the comparative analysis with off-the shelf object detectors yields superior performance of proposed model, measured with standard evaluation strategies. The dataset and the model implementation are made publicly available at this link: https://bit.ly/2TyZICF. [ABSTRACT FROM AUTHOR]

Learning vector embeddings of users and items is the core of modern recommender systems. Recently the collaborative filtering recommender systems based on graph convolutional networks, which integrates the bipartite graph of user-item interaction into the embedding process, has achieved significant success. However, such feature as item-item interaction sequence is neglected in the bipartite graph, which limits the ability to model sequential orders for embedding of items. In this work, we propose a novel item-item interaction sequential graph to globally aggregate the hidden interactions sequence among all items. It is derived from the order of all user-item interactions and can give a supplement for user-item interaction modeling in CF. We also propose an item enhanced graph collaborative network (IEGCN) to mix item-item sequences with user-item interactions for collaborative filtering. We performed experiments on three open datasets, and IEGCN shows substantial improvements in recall and normalized discounted cumulative gain when compared with existing mainstream models. Further analysis verifies the importance of item-item sequence graph to improve the recommendation effect. [ABSTRACT FROM AUTHOR]

Wireless network traffic estimation is of paramount importance for commercial network operators. Accurate traffic forecasting impacts resource allocation that ultimately influences the quality of service perceived by network clients. This kind of forecasting is particularly challenging in real-time scenarios since this operation demands the implementation of advanced artificial intelligence algorithms, which may be prohibitively complex in some scenarios. This article analyzes the trade-off between traffic forecasting accuracy and the complexity of artificial intelligence algorithms applied to this end. Such analysis brings a consistent contribution since the results allow network operators to select the most suitable model for their needs, leading to a well-organized resource sharing initiative that can reduce the amount of investment demanded by these operators to provide a high quality of service to their clients. [ABSTRACT FROM AUTHOR]

Detection of lane region under the road boundary is an imperative module for intelligent vehicle system. Lane markings provide separate regions on the road for the vehicles to avoid the possibility of accidents. Existing methods in lane detection have limited performance using various sensor-based approaches such as Radar and LiDAR and have high operational costs. To achieve a steady and optimal lane detection, the vision-based lane region detection scheme VLDNet is proposed which utilizes a encoder-decoder network using semantic segmentation architecture. In this direction, a hybrid model using UNet and ResNet has been adopted, where UNet is used as a segmentation model and ResNet-50 is used for down-sampling the image and identifying the required features. These identified features have been then applied into UNet for up-sampling and decoding the segments of the images. The publicly available KITTI dataset have been accessed for experiments and validation of the proposed network. The method outperforms the existing state-of-the-art methods in lane region detection. The network achieves better performance using standard evaluation measures such as accuracy of 98.87%, the precision of 98.24%, recall of 96.55%, frequency weighted IoU of 97.78%, and MaxF score of 97.77%. [ABSTRACT FROM AUTHOR]

The demand for several sources of situational data from the traffic environment has intensified in recent years, through the development of applications in intelligent transport systems (ITS), such as autonomous vehicles and advanced driver assistance systems. Among these situational data, the road surface type classification is one of the most important and can be used throughout the ITS domain. However, in order to have a wide application, the development of a safe and reliable model is necessary. Therefore, in addition to the application of safe technology, the model developed must operate correctly in different vehicles, with different driving styles and in different environments in which vehicles can travel to. For this purpose, in this work we collect nine datasets with contextual variations using inertial sensors, represented by accelerometers and gyroscopes. These data were produced in three different vehicles, with three different drivers, in three different environments in which there are three different surface types, in addition to variations in conservation state and presence of obstacles and anomalies, such as speed bumps and potholes. After a pre-processing step, these data were used in 34 different computational models for road surface type classification, employing both Classical Machine Learning and Deep Learning techniques. Through several experiments, we analyze the learning and generalization capacity of each technique. The best model developed was a CNN-based deep neural network, which obtained validation accuracy of 93.17%, classifying surfaces between segments of dirt, cobblestone or asphalt roads. [ABSTRACT FROM AUTHOR]

Attributed network embedding aims at learning low-dimensional network representations in terms of both network structure and attribute information. Most existing methods deal with network structure and attributes separately and combine them in particular ways, which weaken the affinity between structure and attributes and thus lead to suboptimal performance. Moreover, some methods focus solely on local or global network structure, without fully utilizing the structure information underling the network. To address these limitations, we propose structure-guided attributed network embedding with "centroid" enhancement, an unsupervised approach to embed network structure and attribute information comprehensively and seamlessly. Specifically, we regard the neighborhood of each node as a "cluster" and calculate a "centroid" for it through graph convolutional network. We design a "centroid"-based triplet regularizer to impose a gap constraint inspired by K-means. A "centroid"-augment skip-gram model is utilized to deal with high-order proximity. By jointly optimizing the two objectives, the learned representation can preserve both local-global network structure and attribute information. Throughout the model, we exploit network structure to guide the aggregation of attributes, and thus effectively captures the affinity between them. Experimental results on eight real-world datasets demonstrate the superiority of our model over the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

In recent years, neural networks-based autoencoders have gained popularity in problems of anomaly detection. Recent approaches have proposed ensembles of autoencoders to detect network intrusions. The computationally expensive ensembles of autoencoders make it challenging to be used for intrusion detection in networks of devices with lower resources, e.g., the Internet of Things, than in the cloud or data centers. To overcome this challenge, in this work, we propose, investigate and compare four methods to reduce the ensemble complexity through adaptive de-activations of autoencoders. These methods differ in their approach to select the autoencoders to de-activate (criteria-based or random) and differ when they conduct the de-activations (post-training or in-training). Extensive experiments on two recent, realistic IoT intrusion detection datasets validate the effectiveness of the proposed methods in achieving satisfactory detection performance at much lower training, re-training and inference time costs. The proposed methods shall enable scalable and efficient intrusion detection systems or services that could be deployed on-device or on-edge. [ABSTRACT FROM AUTHOR]