Search

With the acceleration of the Internet in Web 2.0, Cloud computing is a new paradigm to offer dynamic, reliable and elastic computing services. Efficient scheduling of resources or optimal allocation of requests is one of the prominent issues in emerging Cloud computing. Considering the growing complexity of Cloud computing, future Cloud systems will require more effective resource management methods. In some complex scenarios with difficulties in directly evaluating the performance of scheduling solutions, classic algorithms (such as heuristics and meta-heuristics) will fail to obtain an effective scheme. Deep reinforcement learning (DRL) is a novel method to solve scheduling problems. Due to the combination of deep learning and reinforcement learning (RL), DRL has achieved considerable performance in current studies. To focus on this direction and analyze the application prospect of DRL in Cloud scheduling, we provide a comprehensive review for DRL-based methods in resource scheduling of Cloud computing. Through the theoretical formulation of scheduling and analysis of RL frameworks, we discuss the advantages of DRL-based methods in Cloud scheduling. We also highlight different challenges and discuss the future directions existing in the DRL-based Cloud scheduling. [ABSTRACT FROM AUTHOR]

Recent studies have shown that deep learning has presented great potential in facial expression recognition (FER) tasks and attracted more and more researchers' attention. Many existing methods have achieved good results on facial expression images in the laboratory environment. However, there are still great challenges for FER in the wild environment where facial expression images are more complex and diverse than those in the laboratory. In this paper, we propose a new method for FER from the perspective of multi-level features extraction and fusion. Different from the existing feature extraction network where only a single convolution kernel scale is present, we propose a feature extraction module with different convolutional kernel scales, which extracts multi-level features as the output of the whole feature extraction network. Further, we do not directly use these multi-level features but propose a feature fusion module with global and local attention to adaptively fuse these different level features in pairs with a top-down way and construct a new facial expression feature. To relieve the overfitting effect caused by data imbalance, we employ label smoothing and L2 regularization strategies to further guide our model forward in a better direction. Through extensive experiments, we demonstrate our method achieves accuracies of 88.08% on RAFDB, 88.11% on FERPlus and 59.38% on AffectNet, respectively, which are very competitive performances. Moreover, our multi-level feature fusion approach enables traditional convolutional backbone networks to improve performance by 0.64–1.13% on FER tasks. [ABSTRACT FROM AUTHOR]

Since its release, the Tensorflow framework has been widely used in various fields due to its advantages in deep learning. However, it is still at its early state. Its native distributed implementation has difficulty in expanding for large models because it has issues of low utilization of multiple GPUs and slow distribution compared with running on single machine. It is of great significance to reduce the training time through parallel models. In view of this, we firstly provided an in-depth analysis of the implementation principle of Tensorflow and identify the bottlenecks of its native distributed parallel models to improve. Then, two optimal algorithms are designed and implemented based on data parallelism and model parallelism modes of Tensorflow. For data parallelism, the proposed algorithm is implemented to replace the native linear execution mode with pipeline execution mode. As for model parallelism, the native random partitioning mode is replaced by our proposed novel greedy algorithm. Finally, we built a homogeneous distributed cluster and a heterogeneous distributed cluster respectively to verify the effectiveness of the proposed algorithms. Through a number of comparative experiments, we showed that the proposed optimal parallel algorithms can effectively reduce model training time by an average of 26.5%(or average 1.5x speedup than native distributed algorithms) and improve the utilization of the cluster while keeping the same accuracy level of native Tensorflow. [ABSTRACT FROM AUTHOR]

With the fast growth of big data applications, it has brought about a huge increase in the energy consumption for big data processing in Cloud data centers. In this study, a frequency-aware and energy-saving strategy based on dynamic voltage and frequency scaling (abbreviated as FAESS-DVFS) is proposed to reduce energy consumption for big data processing in Spark on YARN. Energy saving in two layers (YARN layer and Spark layer) has been designed and implemented for the proposed method. First, an optimal CPU frequency is presented in YARN layer based on the minimum energy efficiency ratio (EER) which can be obtained from status monitoring module. Then, a task scheduling method in Spark layer is constructed to optimize the energy consumption by dynamically adjusting the CPU frequency of nodes in the life cycle of different stages. Test on Hibench, the proposed method can achieve substantial energy saving of up to 29.5% for big data processing compared with the default algorithm in Spark on YARN while satisfying SLA constrains. [ABSTRACT FROM AUTHOR]

Grammatical Error Correction (GEC) is the task of correcting several diverse errors in a text such as spelling, punctuation, morphological, and word choice typos or mistakes. Expressed as a sentence correction task, models such as neural-based sequence-to-sequence (seq2seq) GECs have emerged to offer solutions to the task. However, neural-based seq2seq grammatical error correction models are computationally expensive both in training and in translation inference. Also, they tend to suffer from poor generalization and arrive at inept capabilities due to limited error-corrected data, and thus, incapable of effectively correcting grammar. In this work, we propose the use of Neural Cascading Architecture and different techniques in enhancing the effectiveness of neural sequence-to-sequence grammatical error correction models as inspired by post-editing processes of Neural Machine Translations (NMTs). The findings of our experiments show that, in low-resource NMT models, adapting the presented cascading techniques unleashes performances that is comparable to high setting NMT models, with improvements on state-of-the-art (SOTA) JHU FLuency- Extended GUG corpus (JFLEG) parallel corpus for developing and evaluating GEC model systems. We extensively exploit and evaluate multiple cascading learning strategies and establish best practices toward improving neural seq2seq GECs. [ABSTRACT FROM AUTHOR]

Recent advanced deep learning studies have shown the positive role of feedback mechanism in image super-resolution task. However, current feedback mechanism only calculates residual errors of images with the same resolution without considering the useful features that may be carried by different resolution features. In this paper, to explore the potential of feedback mechanism, we design a new network structure (progressive up- and downsampling back-projection units) to construct a generative adversarial network for single image super-resolution and use progressive growing methodologies to train it. Unlike previous feedback structure, we use progressively increasing scale factor to build up- and down-projection units, which aims to learn fruitful features across scales. This method allows us to get more meaningful information from early feature maps. Additionally, we train our network progressively; in the process of training, we start from single layer network structure and add new layers as the training goes on. By this mean, the training process can be greatly accelerated and stabilized. Experiments on benchmark dataset with the state-of-the-art methods show that our network achieves 0.01 dB, 0.11 dB, 0.13 dB and 0.4 dB better PSNR results than that of RDN+, MDSR, D-DBPN and EDSR on 8 × enlargement, respectively, and also achieves favorable performance against the state-of-the-art methods on 2 × and 4 × enlargement. [ABSTRACT FROM AUTHOR]

Human face recognition have been an active research area for the last few decades. Especially, during the last five years, it has gained significant research attention from multiple domains like computer vision, machine learning and artificial intelligence due to its remarkable progress and broad social applications. The primary goal of any face recognition system is to recognize the human identity from the static images, video data, data-streams and the knowledge of the context in which these data components are being actively used. In this review, we have highlighted major applications, challenges and trends of face recognition systems in social and scientific domains. The prime objective of this research is to sum-up recent face recognition techniques and develop a broad understanding of how these techniques behave on different datasets. Moreover, we discuss some key challenges such as variability in illumination, pose, aging, cosmetics, scale, occlusion, and background. Along with classical face recognition techniques, most recent research directions are deeply investigated, i.e., deep learning, sparse models and fuzzy set theory. Additionally, basic methodologies are briefly discussed, while contemporary research contributions are examined in broader details. Finally, this research presents future aspects of face recognition technologies and its potential significance in the upcoming digital society. [ABSTRACT FROM AUTHOR]

Energy consumption is explosive increasing with the fast growth of big data applications. High carbon emissions from big data platforms have serious impacts on environment. In this paper, we propose an energy-aware scheduling algorithm for Spark (EASAS) to reduce energy consumption while satisfying the service level agreement (SLA). First, we present a new energy consumption model based on Spark framework. Then a strategy table for the relationship between tasks and executors is designed to record the execution time and energy consumption of tasks. The task scheduling in Spark is conducted and optimized based on the strategy table. The proposed strategy overcomes the defect of the default scheduling strategy FIFO and FAIR which cannot perceive energy consumption with the characteristics of energy consumption perception and dynamic optimization scheduling. Compared against FIFO and FAIR, Our EASAS effectively reduces on average about 25–40% of the total energy consumption of Spark applications under deadline constrains. [ABSTRACT FROM AUTHOR]

3D-R2N2 and other advanced 3D reconstruction neural networks have achieved impressive results, however most of them still suffer from training difficulties and detail losing, due to their weak feature extraction capability and improper loss function. This paper aims to overcome these shortcomings and defects by building a brand new model based on 3D-R2N2. The new model adopts densely connected structure as encoder, and utilizes Chamfer Distance as loss function. The aim is to enhance the learning ability of the network for complex data, meanwhile, make the focus of the whole network rest on the reconstruction of detail structures. In addition, we also made an improved decoder by building two parallel predictor branches to make better use of the feature information and boost the network's performance on reconstruction task. Through extensive tests, the results show that our proposed model called 3D-R2N2-V2 is slightly slower than 3D-R2N2 in predicting speed, but it can be 20% to 30% faster than 3D-R2N2 in training speed and obtain 15% and 10% better voxel IoU results on both single- and multi-view reconstruction tasks, respectively. Compared with other recent state-of-the-art methods like OGN and DRC, the reconstruction effect of our approach is also competitive. [ABSTRACT FROM AUTHOR]

In virtualized data centers, consolidation of virtual machines (VMs) on minimizing the number of total physical machines (PMs) has been recognized as a very efficient approach. This paper considers the energy-efficient consolidation of VMs in a cloud datacenter. Concentrating on CPU-intensive applications, the objective is to schedule all requests non-preemptively, subjecting to constraints of PM capacities and running time interval spans, to make the total energy consumption of all PMs is minimized (called MinTE for abbreviation). The MinTE problem is NP-complete in general. We propose a self-adaptive approach called SAVE. The approach makes decisions of the assignment and migration of VMs by probabilistic processes and is based exclusively on local information. Both simulation and real environment test show that our proposed method SAVE can reduce energy consumption about 30 % against VMWare DRS and 10–20% against ecoCloud on average. Extensive experiments show that our method outperforms the existing method and achieves significant energy savings and high utilization. [ABSTRACT FROM AUTHOR]

High energy consumption in data center has become a key problem, and the Carbon (CO $$_2$$ ) emissions from data center have serious impacts on environment. To improve traditional performance-energy model and minimize energy consumption in cloud data center, an energy-efficient and QoS-aware (EEQoS) model is developed for virtual resources consolidation. The EEQoS model is designed to investigate the trade-off between energy and QoS. The user's satisfaction degree and Euclidean distance of resources are also respectively applied to the QoS model and power model. In addition, this EEQoS model is combined with the particle swarm optimization algorithm by setting the power consumption per QoS value as the objective function to consolidate virtual resources in data center. Compared with traditional power model, the proposed model reduces response time by 27.2%, cost by 31.4%, SLA violations by 40.5% and improves throughput by 13.5% on average while increasing less than 3.8% energy consumption during the whole test period. [ABSTRACT FROM AUTHOR]

Large-scale MapReduce clusters that routinely process big data bring challenges to the cloud computing. One of the key challenges is to reduce the response time of these MapReduce clusters by minimizing their makespans. It is observed that the order in which these jobs are executed can have a significant impact on their overall makespans and resource utilization. In this work, we consider a scheduling model for multiple MapReduce jobs. The goal is to design a job scheduler that minimizes the makespan of such a set of MapReduce jobs. We exploit classical Johnson model and propose a novel framework HScheduler, which combines features of both classical Johnson's algorithm and MapReduce to minimize the makespan for both offline and online jobs. Our Offline HScheduler reaches the theoretical lower bound (optimum) and Online HScheduler is 2-competitive which is the best-known constant ratio for minimizing the makespan. Through extensive real data tests, we find that HScheduler has better performance than the best-known approach by 10.6-11.7 % on average for offline scheduling and 8-10 % on average for online scheduling. The HScheduler can be applied to improve responsive time, throughput and energy efficiency in cloud computing. [ABSTRACT FROM AUTHOR]

Recent studies show that both data traffic and real-time traffic grow very fast in wired and wireless networks. To provide better performance guarantee, these applications need efficient network modeling and planning. In this paper, the problem where the total bandwidth of a link is shared by streaming traffic (real time traffic such as voice or video etc.) and elastic traffic (such as data) is studied. Integrating streaming traffic and elastic traffic presents a unique dimensioning problem. This paper considers dimensioning a link to satisfy both quality of service (QoS) requirements for streaming traffic, such as loss probability, and elastic traffic, such as mean waiting (delay) time. The Erlang loss model is applied to streaming traffic and a bursty traffic model is applied to the elastic traffic. Efficient dimensioning algorithms based on classical Markovian models and time-scale decomposition are then proposed. Numerical results show that the proposed methods have good accuracy. [ABSTRACT FROM AUTHOR]

A distributed coevolutionary algorithm is proposed to solve the multiobjective hybrid flowshop scheduling problems to minimize the maximum completion time and total tardiness of jobs. The framework of the distributed coevolutionary algorithm consists of a global agent and multiple local agents. The global agent and local agents evolve independently and cooperate by interchanging a selected solution list. Unlike the cooperative coevolutionary algorithms in the literature, the proposed algorithm does not decompose the scheduling problem and executes evolutionary operations based on the whole solution of the problem in all the agents. SPEA2 is the core components in the local agents. Path relinking is applied in order to implement the evolutionary computation among non-dominated solutions in the global agent. We analyzed the time complexity of the proposed algorithm. To evaluate the performance against the benchmark of multiobjective evolutionary algorithms, it is tested on a large number of computational instances. The computational experiments show the proposed distributed coevolutionary algorithm can obtain better solution quality than other algorithms within given computational time. [ABSTRACT FROM AUTHOR]

This paper considers online energy-efficient scheduling of virtual machines (VMs) for Cloud data centers. Each request is associated with a start-time, an end-time, a processing time and a capacity demand from a Physical Machine (PM). The goal is to schedule all of the requests non-preemptively in their start-time-end-time windows, subjecting to PM capacity constraints, such that the total busy time of all used PMs is minimized (called MinTBT-ON for abbreviation). This problem is a fundamental scheduling problem for parallel jobs allocation on multiple machines; it has important applications in power-aware scheduling in cloud computing, optical network design, customer service systems, and other related areas. Offline scheduling to minimize busy time is NP-hard already in the special case where all jobs have the same processing time and can be scheduled in a fixed time interval. One best-known result for MinTBT-ON problem is a g-competitive algorithm for general instances and unit-size jobs using First-Fit algorithm where g is the total capacity of a machine. In this paper, a $(1+\frac{g-2}{k}-\frac{g-1}{k^{2}})$ -competitive algorithm, Dynamic Bipartition-First-Fit (BFF) is proposed and proved for general case, where k is the ratio of the length of the longest interval over the length of the second longest interval for k>1 and g≥2. More results in general and special cases are obtained to improve the best-known bounds. [ABSTRACT FROM AUTHOR]

Technology for monitoring in vivo microRNA (miRNA) activity is extremely important for elucidating miRNA biology. However, in vivo studies of miRNA have been hampered by the lack of a convenient approach to reliably reflect real-time functional changes in miRNAs. Sensors for miRNA were developed by adding miRNA target sequences to the 3′-untranslated region of Gaussia princeps luciferase ( Gluc) mRNA. These sensors were then evaluated in vitro and in vivo by measuring Gluc activity in cell supernatants and in peripheral blood. Sensors driven by the CMV promoter were effective for monitoring miR-122 in living cells, but not for the long-term monitoring of miR-122 or miR-142 in mouse liver because of CMV-promoter silencing. Replacing the CMV promoter with a CAG promoter rendered these sensors effective for the long-term monitoring of relevant liver miRNA activities. We subsequently used the CAG-promoter-based sensor for the long-term monitoring of endogenous liver miR-122, miR142 and miR-34a activities, as well as for exogenous miR-34a activity. Our study demonstrates that real-time in vivo activities of miRNAs can be continuously and conveniently detected in mouse liver using the sensors that we have developed. [ABSTRACT FROM AUTHOR]