Search

Linear spectral unmixing is the practice of decomposing the mixed pixel into a linear combination of the constituent endmembers and the estimated abundances. This paper focuses on unsupervised spectral unmixing where the endmembers are unknown a priori. Conventional approaches use either geometrical- or statistical-based approaches. In this paper, we address the challenges of spectral unmixing with unsupervised deep learning models, in specific, the autoencoder models, where the decoder serves as the endmembers and the hidden layer output serves as the abundances. In several recent attempts, part-based autoencoders have been designed to solve the unsupervised spectral unmixing problem. However, the performance has not been satisfactory. In this paper, we first discuss some important findings we make on issues with part-based autoencoders. By proof of counterexample, we show that all existing part-based autoencoder networks with nonnegative and tied encoder and decoder are inherently defective by making these inappropriate assumptions on the network structure. As a result, they are not suitable for solving the spectral unmixing problem. We propose a so-called untied denoising autoencoder with sparsity, in which the encoder and decoder of the network are independent, and only the decoder of the network is enforced to be nonnegative. Furthermore, we make two critical additions to the network design. First, since denoising is an essential step for spectral unmixing, we propose to incorporate the denoising capacity into the network optimization in the format of a denoising constraint rather than cascading another denoising preprocessor in order to avoid the introduction of additional reconstruction error. Second, to be more robust to the inaccurate estimation of a number of endmembers, we adopt an $l_{21}$ -norm on the encoder of the network to reduce the redundant endmembers while decreasing the reconstruction error simultaneously. The experimental results demonstrate that the proposed approach outperforms several state-of-the-art methods, especially for highly noisy data. [ABSTRACT FROM AUTHOR]

This paper investigates the stochastic optimization problem focusing on developing scalable parallel algorithms for deep learning tasks. Our solution involves a reformation of the objective function for stochastic optimization in neural network models, along with a novel parallel computing strategy, coined the weighted aggregating stochastic gradient descent (WASGD). Following a theoretical analysis on the characteristics of the new objective function, WASGD introduces a decentralized weighted aggregating scheme based on the performance of local workers. Without any center variable, the new method automatically gauges the importance of local workers and accepts them by their contributions. Furthermore, we have developed an enhanced version of the method, WASGD+, by (1) implementing a designed sample order and (2) upgrading the weight evaluation function. To validate the new method, we benchmark our pipeline against several popular algorithms including the state-of-the-art deep neural network classifier training techniques (e.g., elastic averaging SGD). Comprehensive validation studies have been conducted on four classic datasets: CIFAR-100, CIFAR-10, Fashion-MNIST, and MNIST. Subsequent results have firmly validated the superiority of the WASGD scheme in accelerating the training of deep architecture. Better still, the enhanced version, WASGD+, is shown to be a significant improvement over its prototype. [ABSTRACT FROM AUTHOR]

For vehicle-to-network communications, handover (HO) management enables vehicles to maintain the connection with the network while transiting through coverage areas of different base stations (BSs). However, the high mobility of vehicles means shorter connection periods with each BS that leads to frequent HOs, hence raises the necessity for optimal HO decision making for high quality infotainment services. Machine learning is capable of capturing underlying pattern via data driven methods to find optimal solutions to complex problems, and much learning-based HO optimization research has been conducted focusing on specific network setups. However, attention still needs to be paid to the actual deployment aspect and standardized datasets or simulation environments for evaluation. This paper proposes a deep reinforcement learning-based HO algorithm using the input parameters that are configurable in the existing measurement report of cellular networks. The performance of the proposed algorithm is evaluated using the well-known network simulator ns-3 with its official LTE module. A realistic network setup in the city center of Glasgow (U.K.) is configured with vehicle trajectories generated by the routes mobility model using Google Maps Directions API. Evaluation results reveal that the proposed algorithm significantly outperforms the A3 RSRP baseline with an average of 25.72% packet loss reduction per HO, suggesting significant improvement in quality of service of phone call and video streaming, etc. The proposed algorithm also has a small implementation cost compared to some state-of-the-art and should be deployed by a software update to a local BS controller. [ABSTRACT FROM AUTHOR]

Neuro-inspired architectures based on synaptic memory arrays have been proposed for on-chip acceleration of weighted sum and weight update in machine/deep learning algorithms. In this paper, we developed NeuroSim, a circuit-level macro model that estimates the area, latency, dynamic energy, and leakage power to facilitate the design space exploration of neuro-inspired architectures with mainstream and emerging device technologies. NeuroSim provides flexible interface and a wide variety of design options at the circuit and device level. Therefore, NeuroSim can be used by neural networks (NNs) as a supporting tool to provide circuit-level performance evaluation. With NeuroSim, an integrated framework can be built with hierarchical organization from the device level (synaptic device properties) to the circuit level (array architectures) and then to the algorithm level (NN topology), enabling instruction-accurate evaluation on the learning accuracy as well as the circuit-level performance metrics at the run-time of online learning. Using multilayer perceptron as a case-study algorithm, we investigated the impact of the “analog” emerging nonvolatile memory (eNVM)’s “nonideal” device properties and benchmarked the tradeoffs between SRAM, digital, and analog eNVM-based architectures for online learning and offline classification. [ABSTRACT FROM AUTHOR]

Unmanned aerial vehicle (UAV) technology is recognized as a promising solution to area coverage problems (ACPs) and has been extensively studied recently. In this paper, we study the 3D irregular terrain surface coverage problem with a hierarchical UAV swarm. We first build the 3D model of a random irregular terrain and propose a geometric way to project the 3D terrain surface into many weighted 2D patches. Then we develop a two-level hierarchical UAV swarm architecture, including the low-level follower UAVs (FUAVs) and the high-level leader UAVs (LUAVs). For FUAVs, we design a coverage trajectory algorithm to carry out specific coverage tasks within patches based on the star communication topology. For LUAVs, we propose a swarm deep Q-learning (SDQN) reinforcement learning algorithm to select patches. Moreover, an observation history model based on convolutional neural networks (CNNs) and the mean embedding method is integrated into SDQN to address the communication limitation problems of LUAVs. The numerical results show that FUAVs can cover the entire area of each patch with little redundancies, and the total coverage time of the SDQN is less than that of existing methods, which demonstrates the effectiveness of the proposed algorithms. [ABSTRACT FROM AUTHOR]

Recent 5G trials have demonstrated the usefulness of the Network Slicing concept that delivers customizable services to new and under-serviced industry sectors. However, user mobility’s impact on the optimal resource allocation within and between slices deserves more attention. Slices and their dedicated resources should be offered where the services are to be consumed to minimize network latency and associated overheads and costs. Different mobility patterns lead to different resource re-allocation triggers, leading eventually to slice mobility when enough resources are to be migrated. The selection of the proper triggers for resource re-allocation and related slice mobility patterns is challenging due to triggers’ multiplicity and overlapping nature. In this paper, we investigate the applicability of two Deep Reinforcement Learning based algorithms for allowing a fine-grained selection of mobility triggers that may instantiate slice and resource mobility actions. While the first proposed algorithm relies on a value-based learning method, the second one exploits a hybrid approach to optimize the action selection process. We present an enhanced ETSI Network Function Virtualization edge computing architecture that incorporates the studied mechanisms to implement service and slice migration. We evaluate the proposed methods’ efficiency in a simulated environment and compare their performance in terms of training stability, learning time, and scalability. Finally, we identify and quantify the applicability aspects of the respective approaches. [ABSTRACT FROM AUTHOR]

In this paper, we seek to improve deep neural networks by generalizing the pooling operations that play a central role in the current architectures. We pursue a careful exploration of approaches to allow pooling to learn and to adapt to complex and variable patterns. The two primary directions lie in: (1) learning a pooling function via (two strategies of) combining of max and average pooling, and (2) learning a pooling function in the form of a tree-structured fusion of pooling filters that are themselves learned. In our experiments every generalized pooling operation we explore improves performance when used in place of average or max pooling. We experimentally demonstrate that the proposed pooling operations provide a boost in invariance properties relative to conventional pooling and set the state of the art on several widely adopted benchmark datasets. These benefits come with only a light increase in computational overhead during training (ranging from additional 5 to 15 percent in time complexity) and a very modest increase in the number of model parameters (e.g., additional 1, 9, and 27 parameters for mixed, gated, and 2-level tree pooling operators, respectively). To gain more insights about our proposed pooling methods, we also visualize the learned pooling masks and the embeddings of the internal feature responses for different pooling operations. Our proposed pooling operations are easy to implement and can be applied within various deep neural network architectures. [ABSTRACT FROM PUBLISHER]

Autoencoders are deep learning architectures that learn feature representation by minimizing the reconstruction error. Using an autoencoder as baseline, this paper presents a novel formulation for a class sparsity based supervised encoder, termed as CSSE. We postulate that features from the same class will have a common sparsity pattern/support in the latent space. Therefore, in the formulation of the autoencoder, a supervision penalty is introduced as a joint-sparsity promoting l2,1 -norm. The formulation of CSSE is derived for a single hidden layer and it is applied for multiple hidden layers using a greedy layer-by-layer learning approach. The proposed CSSE approach is applied for learning face representation and verification experiments are performed on the LFW and PaSC face databases. The experiments show that the proposed approach yields improved results compared to autoencoders and comparable results with state-of-the-art face recognition algorithms. [ABSTRACT FROM AUTHOR]

In the current unmanned aircraft systems (UASs) for sensing services, unmanned aerial vehicles (UAVs) transmit their sensory data to terrestrial mobile devices over the unlicensed spectrum. However, the interference from surrounding terminals is uncontrollable due to the opportunistic channel access. In this paper, we consider a cellular Internet of UAVs to guarantee the Quality-of-Service (QoS), where the sensory data can be transmitted to the mobile devices either by UAV-to-Device (U2D) communications over cellular networks, or directly through the base station (BS). Since UAVs’ sensing and transmission may influence their trajectories, we study the trajectory design problem for UAVs in consideration of their sensing and transmission. This is a Markov decision problem (MDP) with a large state-action space, and thus, we utilize multi-agent deep reinforcement learning (DRL) to approximate the state-action space, and then propose a multi-UAV trajectory design algorithm to solve this problem. Simulation results show that our proposed algorithm can achieve a higher total utility than policy gradient algorithm and single-agent algorithm. [ABSTRACT FROM AUTHOR]

Virtual network embedding arranges virtual network services onto substrate network components. The performance of embedding algorithms determines the effectiveness and efficiency of a virtualized network, making it a critical part of the network virtualization technology. To achieve better performance, the algorithm needs to automatically detect the network status which is complicated and changes in a time-varying manner, and to dynamically provide solutions that can best fit the current network status. However, most existing algorithms fail to provide automatic embedding solutions in an acceptable running time. In this paper, we combine deep reinforcement learning with a novel neural network structure based on graph convolutional networks, and propose a new and efficient algorithm for automatic virtual network embedding. In addition, a parallel reinforcement learning framework is used in training along with a newly-designed multi-objective reward function, which has proven beneficial to the proposed algorithm for automatic embedding of virtual networks. Extensive simulation results under different scenarios show that our algorithm achieves best performance on most metrics compared with the existing state-of-the-art solutions, with upto 39.6% and 70.6% improvement on acceptance ratio and average revenue, respectively. Moreover, the results also demonstrate that the proposed solution possesses good robustness. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel sparse signal recovery algorithm called the trainable iterative soft thresholding algorithm (TISTA). The proposed algorithm consists of two estimation units: a linear estimation unit and a minimum mean squared error (MMSE) estimator based shrinkage unit. The error variance required in the MMSE shrinkage unit is precisely estimated from a tentative estimate of the original signal. The remarkable feature of the proposed scheme is that TISTA includes adjustable variables that control step size and the error variance for the MMSE shrinkage function. The variables are adjusted by standard deep learning techniques. The number of trainable variables of TISTA is nearly equal to the number of iteration rounds and is much smaller than that of known learnable sparse signal recovery algorithms. This feature leads to highly stable and fast training processes of TISTA. Computer experiments show that TISTA is applicable to various classes of sensing matrices, such as Gaussian matrices, binary matrices, and matrices with large condition numbers. Numerical results also demonstrate that, in many cases, TISTA provides significantly faster convergence than approximate message passing (AMP) and the learned iterative shrinkage thresholding algorithm and also outperforms orthogonal AMP in the NMSE performance. [ABSTRACT FROM AUTHOR]

Compressive sensing (CS) has proved effective for tomographic reconstruction from sparsely collected data or under-sampled measurements, which are practically important for few-view computed tomography (CT), tomosynthesis, interior tomography, and so on. To perform sparse-data CT, the iterative reconstruction commonly uses regularizers in the CS framework. Currently, how to choose the parameters adaptively for regularization is a major open problem. In this paper, inspired by the idea of machine learning especially deep learning, we unfold the state-of-the-art “fields of experts”-based iterative reconstruction scheme up to a number of iterations for data-driven training, construct a learned experts’ assessment-based reconstruction network (LEARN) for sparse-data CT, and demonstrate the feasibility and merits of our LEARN network. The experimental results with our proposed LEARN network produces a superior performance with the well-known Mayo Clinic low-dose challenge data set relative to the several state-of-the-art methods, in terms of artifact reduction, feature preservation, and computational speed. This is consistent to our insight that because all the regularization terms and parameters used in the iterative reconstruction are now learned from the training data, our LEARN network utilizes application-oriented knowledge more effectively and recovers underlying images more favorably than competing algorithms. Also, the number of layers in the LEARN network is only 50, reducing the computational complexity of typical iterative algorithms by orders of magnitude. [ABSTRACT FROM AUTHOR]

Model-based iterative reconstruction algorithms for low-dose X-ray computed tomography (CT) are computationally expensive. To address this problem, we recently proposed a deep convolutional neural network (CNN) for low-dose X-ray CT and won the second place in 2016 AAPM Low-Dose CT Grand Challenge. However, some of the textures were not fully recovered. To address this problem, here we propose a novel framelet-based denoising algorithm using wavelet residual network which synergistically combines the expressive power of deep learning and the performance guarantee from the framelet-based denoising algorithms. The new algorithms were inspired by the recent interpretation of the deep CNN as a cascaded convolution framelet signal representation. Extensive experimental results confirm that the proposed networks have significantly improved performance and preserve the detail texture of the original images. [ABSTRACT FROM AUTHOR]

Following the great success of deep convolutional neural networks (CNNs) in computer vision, this paper proposes a complex-valued CNN (CV-CNN) specifically for synthetic aperture radar (SAR) image interpretation. It utilizes both amplitude and phase information of complex SAR imagery. All elements of CNN including input-output layer, convolution layer, activation function, and pooling layer are extended to the complex domain. Moreover, a complex backpropagation algorithm based on stochastic gradient descent is derived for CV-CNN training. The proposed CV-CNN is then tested on the typical polarimetric SAR image classification task which classifies each pixel into known terrain types via supervised training. Experiments with the benchmark data sets of Flevoland and Oberpfaffenhofen show that the classification error can be further reduced if employing CV-CNN instead of conventional real-valued CNN with the same degrees of freedom. The performance of CV-CNN is comparable to that of existing state-of-the-art methods in terms of overall classification accuracy. [ABSTRACT FROM PUBLISHER]

This article presents a deep learning-based classifier IC for keyword spotting (KWS) in 65-nm CMOS designed using an algorithm-hardware co-design approach. First, a recurrent attention model (RAM) algorithm for the KWS task (the KeyRAM algorithm) is proposed. The KeyRAM algorithm enables accuracy versus energy scalability via a confidence-based computation (CC) scheme, leading to a $2.5\times $ reduction in computational complexity compared to state-of-the-art (SOTA) neural networks, and is well-suited for in-memory computing (IMC) since the bulk (89%) of its computations are 4-b matrix-vector multiplies. The KeyRAM IC comprises a multi-bit multi-bank IMC architecture with a digital co-processor. A sparsity-aware summation scheme is proposed to alleviate the challenge faced by IMCs when summing sparse activations. The digital co-processor employs diagonal major weight storage to compute without any stalls. This combination of the IMC and digital processors enables a balanced tradeoff between energy efficiency and high accuracy computation. The resultant KWS IC achieves SOTA decision latency of 39.9 $\mu \text{s}$ with a decision energy < 0.5 $\mu \text{J}$ /dec which translates to more than 24 $\times $ savings in the energy-delay product (EDP) of decisions over existing KWS ICs. [ABSTRACT FROM AUTHOR]

Depth image super-resolution is a challenging problem, since normally high upscaling factors are required (e.g., $16\times $), and depth images are often noisy. In order to achieve large upscaling factors and resilience to noise, we propose a Robust Algorithm for Depth imAge super Resolution (RADAR) that combines the power of finite rate of innovation (FRI) theory with multimodal dictionary learning. Given a low-resolution (LR) depth image, we first model its rows and columns as piece-wise polynomials and propose an FRI-based depth upscaling (FDU) algorithm to super-resolve the image. Then, the upscaled moderate quality (MQ) depth image is further enhanced with the guidance of a registered high-resolution (HR) intensity image. This is achieved by learning multimodal mappings from the joint MQ depth and HR intensity pairs to the HR depth, through a recently proposed triple dictionary learning (TDL) algorithm. Moreover, to speed up the super-resolution process, we introduce a new projection-based rapid upscaling (PRU) technique that pre-calculates the projections from the joint MQ depth and HR intensity pairs to the HR depth. Compared with the state-of-the-art deep learning-based methods, our approach has two distinct advantages: we need a fraction of training data but can achieve the best performance, and we are resilient to mismatches between training and testing datasets. The extensive numerical results show that the proposed method outperforms other state-of-the-art methods on either noise-free or noisy datasets with large upscaling factors up to $16\times $ and can handle unknown blurring kernels well. [ABSTRACT FROM AUTHOR]

Machine learning (ML) models have been used to improve the quality of different physical design steps, such as timing analysis, clock tree synthesis, and routing. However, so far very few works have addressed the problem of algorithm selection during physical design, which can drastically reduce the computational effort of some steps. This work proposes a legalization algorithm selection framework using deep convolutional neural networks (CNNs). To extract features, we used snapshots of circuit placements and used transfer learning to train the models using pretrained weights of the Squeezenet architecture. By doing so, we can greatly reduce the training time and required data even though the pretrained weights come from a different problem. We performed extensive experimental analysis of ML models, providing details on how we chose the parameters of our model, such as CNN architecture, learning rate, and number of epochs. We evaluated the proposed framework by training a model to select between different legalization algorithms according to cell displacement and wirelength variation. The trained models achieved an average $F$ -score of 0.98 when predicting cell displacement and 0.83 when predicting wirelength variation. When integrated into the physical design flow, the cell displacement model achieved the best results on 15 out of 16 designs, while the wirelength variation model achieved that for 10 out of 16 designs, being better than any individual legalization algorithm. Finally, using the proposed ML model for algorithm selection resulted in a speedup of up to $10\times $ compared to running all the algorithms separately. [ABSTRACT FROM AUTHOR]

For the large amount of spatial and spectral information contained in hyperspectral image (HSI), feature description of HSI has attracted widespread concern in recent years. Existing deep learning-based HSI feature description algorithms require a large number of training samples and have poor interpretability. Therefore, it is necessary to develop an efficient HSI features description algorithm with interpretability based on machine learning. Local binary pattern (LBP) is a classical descriptor used to extract the local spatial texture features of images, which has been widely applied to image feature description and matching. However, the existing LBP algorithms for HSI are based on the single-dimensional description, which leads to the limitations on the expression of spatial–spectral information. Therefore, a multidimensional LBP (MDLBP) based on Clifford algebra for HSI is proposed in this article, which is able to extract spatial–spectral feature from multiple dimensions. First, with the theory of the Clifford algebra, a new representation of HSI including spatial and spectral information is built. Second, the geometric relationship between the local geometry of HSI in Clifford algebra space is calculated to realize the local multidimensional description of the local spatial–spectral information. Finally, a novel LBP coding algorithm for HSI is implemented based on the local multidimensional description to calculate the feature descriptor of HSI. The experimental results on HSI classification show that our proposed MDLBP algorithm can achieve higher accuracy than the representative spatial–spectral features and the existing LBP algorithms, especially in the scenery of small-scale training samples. [ABSTRACT FROM AUTHOR]

In the backpropagation algorithm, the error calculated from the output of the neural network should backpropagate the layers to update the weights of each layer, making it difficult to parallelize the training process and requiring frequent off-chip memory access. Local learning algorithms locally generate error signals which are used for weight updates, removing the need for backpropagation of error signals. However, prior works rely on large, complex auxiliary networks for reliable training, which results in large computational overhead and undermines the advantages of local learning. In this work, we propose a local learning algorithm that significantly reduces computational complexity as well as improves training performance. Our algorithm combines multiple consecutive layers into a block and performs local learning on a block-by-block basis, while dynamically changing block boundaries during training. In experiments, our approach achieves 95.68% and 79.42% test accuracy on the CIFAR-10 and CIFAR-100 datasets, respectively, using a small fully connected layer as auxiliary networks, closely matching the performance of the backpropagation algorithm. Multiply-accumulate (MAC) operations and off-chip memory access also reduce by up to 15% and 81% compared to backpropagation. [ABSTRACT FROM AUTHOR]

Fault diagnosis of rolling bearings is an essential process for improving the reliability and safety of the rotating machinery. It is always a major challenge to ensure fault diagnosis accuracy in particular under severe working conditions. In this article, a deep adversarial domain adaptation (DADA) model is proposed for rolling bearing fault diagnosis. This model constructs an adversarial adaptation network to solve the commonly encountered problem in numerous real applications: the source domain and the target domain are inconsistent in their distribution. First, a deep stack autoencoder (DSAE) is combined with representative feature learning for dimensionality reduction, and such a combination provides an unsupervised learning method to effectively acquire fault features. Meanwhile, domain adaptation and recognition classification are implemented using a Softmax classifier to augment classification accuracy. Second, the effects of the number of hidden layers in the stack autoencoder network, the number of neurons in each hidden layer, and the hyperparameters of the proposed fault diagnosis algorithm are analyzed. Third, comprehensive analysis is performed on real data to validate the performance of the proposed method; the experimental results demonstrate that the new method outperforms the existing machine learning and deep learning methods, in terms of classification accuracy and generalization ability. [ABSTRACT FROM AUTHOR]

While many approaches exist in the literature to learn low-dimensional representations for data collections in multiple modalities, the generalizability of multi-modal nonlinear embeddings to previously unseen data is a rather overlooked subject. In this work, we first present a theoretical analysis of learning multi-modal nonlinear embeddings in a supervised setting. Our performance bounds indicate that for successful generalization in multi-modal classification and retrieval problems, the regularity of the interpolation functions extending the embedding to the whole data space is as important as the between-class separation and cross-modal alignment criteria. We then propose a multi-modal nonlinear representation learning algorithm that is motivated by these theoretical findings, where the embeddings of the training samples are optimized jointly with the Lipschitz regularity of the interpolators. Experimental comparison to recent multi-modal and single-modal learning algorithms suggests that the proposed method yields promising performance in multi-modal image classification and cross-modal image-text retrieval applications. [ABSTRACT FROM AUTHOR]

Through well-designed optimization paradigm and deep neural networks as feature extractor, deep reinforcement learning (DRL) algorithms learn optimal policy on discrete and continuous action space. However, such capability is restricted by the low sampling efficiency. By inspecting the importance of feature extraction in DRL, we find that state feature learning is one of the key obstacles for sampling efficiently. To this end, we propose a new state representation learning scheme with adjacent state consistency loss (ASC loss). The loss is based on the hypothesis that the distance between adjacent states is smaller than that of far apart ones since scenes in videos generally evolve smoothly. We exploit ASC loss as an assistant of RL loss in the training phase to boost the state feature learning, and make evaluation on existing DRL algorithms as well as behavioral cloning algorithm. Experiments on Atari games and MuJoCo continuous control tasks demonstrate the effectiveness of our scheme. [ABSTRACT FROM AUTHOR]

Nowadays, large and complex deep learning (DL) models are increasingly trained in a distributed manner across multiple worker machines, in which extensive communications between workers pose serious scaling problems. In this article, we present a quantitative survey of communication optimization techniques for data parallel distributed DL. We first identify the major communication challenges and classify the existing solutions into three levels, namely the learning algorithm, the system architecture, and the network infrastructure. We present the state-of-the-art communication optimization techniques and conduct a comparative study of seven common lossless distributed DL methods on a 32-GPU cluster with 100Gb/s InfiniBand (IB). We show that the DL models with low model intensity (such as BERT and BERT-Large) are difficult to scale out even with the best available lossless algorithm over 100Gb/s IB; and the system architecture and scheduling algorithms have a critical impact on the scaling property. We conclude the article with discussions of open issues for further investigation. [ABSTRACT FROM AUTHOR]

Aiming at the fault states of the split pins in the catenary fastening devices of high-speed railway, such as missing, loosening, and improper installation, a method for detecting the defects in the split pins is proposed. First, the split pins are localized by a two-stage positioning method based on the improved YOLOV3 algorithm. The first stage is used to localize five joint components on the catenary support devices and the second stage is applied to locate the split pins in the joint component images. Then, the deeplabv3+ algorithm is implemented for semantic segmentation on the split-pin images. Finally, the split pins are classified based on the semantic segmentation images, in accordance with the split-pins’ semantic information of the head, body, and tail. In order to verify the adaptability and accuracy of the proposed method, the catenary support device images in the multihigh-speed railway lines and multienvironments were tested. Meanwhile, our algorithm is compared with other deep learning algorithm. The results show that the proposed method has a higher accuracy in detecting the defects in the split pins, which can guarantee the stable operation of the catenary support devices. [ABSTRACT FROM AUTHOR]

Electrocardiogram (ECG) is often used together with a spectral Doppler ultrasound to separate heart cycles by determining the end-diastole locations. However, the ECG signal is not always recorded. In such cases, the cardiac cycles can be estimated manually from the ultrasound data retrospectively. We present a deep learning-based method for automatic detection of the end-diastoles in spectral Doppler spectrograms. The method uses a combination of a convolutional neural network (CNN) for extracting features and a recurrent neural network (RNN) for modeling temporal relations. In echocardiography, there are three Doppler spectrogram modalities, continuous wave, pulsed wave, and tissue velocity Doppler. Both the training and test data sets include all three modalities. The model was tested on 643 spectrograms coming from different hospitals than in the training data set. For the purposes described in this work, a valid end-diastole detection is defined as a prediction being closer than 60 ms to the reference value. We will refer to these as true detections. Similarly, a prediction farther away is defined as nonvalid or false detections. The method automatically rejects spectrograms where the detection of an end-diastole has low confidence. When setting the algorithm to reject 1.9%, the method achieved 97.7% true detections with a mean error of 14 ms and had 2.5% false detections on the remaining spectrograms. [ABSTRACT FROM AUTHOR]

With the exponential growth of data and the high demand for the analysis of large datasets, the MapReduce framework has been widely utilized to process data in a timely, cost-effective manner. It is well-known that the performance of MapReduce is limited by its default configuration parameters, and there are a few research studies that have focused on finding the optimal configurations to improve the performance of the MapReduce framework. Recently, machine learning based approaches have been receiving more attention to be utilized to auto configure the MapReduce parameters to account for the dynamic nature of the applications. In this article, we propose and develop a reinforcement learning (RL)-based scheme, named RL-MRCONF, to automatically configure the MapReduce parameters. Specifically, we explore and experiment with two variations of RL-MRCONF; one variation is based on the traditional RL algorithm and the second is based on the deep RL algorithm. Results obtained from simulations show that the RL-MRCONF has the ability to successfully and effectively auto-configure the MapReduce parameters dynamically according to changes in job types and computing resources. Moreover, simulation results show our proposed RL-MRCONF scheme outperforms the traditional RL-based implementation. Using datasets provided by MR-Perf, simulation results show that our proposed scheme provides around 50% performance improvement in terms of execution time when compared with MapReduce using default settings. [ABSTRACT FROM AUTHOR]

Power and thermal management are critical components of high-performance-computing (HPC) systems, due to their high-power density and large total power consumption. The assessment of thermal dissipation by means of compact models directly from the thermal response of the final device enables more robust and precise thermal control strategies as well as automated diagnosis. However, when dealing with large-scale systems “in production,” the accuracy of learned thermal models depends on the dynamics of the power excitation, which depends also on the executed workload, and measurement nonidealities such as quantization. In this article we show that, using an advanced system identification algorithm, we are able to generate very accurate thermal models (average error lower than our sensors quantization step of 1 °C) for a large-scale HPC system on real workloads for very long time periods. However, we also show that: 1) not all real workloads allow for the identification of a good model and 2) starting from the theory of system identification it is very difficult to evaluate if a trace of data leads to a good estimated model. We then propose and validate a set of techniques based on machine learning and deep learning algorithms for the choice of data traces to be used for model identification. We also show that deep learning techniques are absolutely necessary to correctly choose such traces up to 96% of the times. [ABSTRACT FROM AUTHOR]

For a battery swapping station (BSS), the stochastic operation of electric buses (EBs) and the uncertainty of electricity prices cause unnecessary economic losses. To minimize the operating costs of the BSS, this article applies the deep reinforcement learning (DRL) and proposes a BSS model to determine the optimal real-time charge/discharge power of the charging piles. The predictability of bus operation and the uncertainty of price can be directly captured from historical data without any assumption in the model. Moreover, deep deterministic policy gradient (DDPG), as the DRL algorithm, is implemented in the model to simultaneously control multiple charging piles. Numerical results illustrate that the proposed approach can bring less operating cost than the existing benchmark control methods for a BSS while providing adequate batteries ready for swapping. [ABSTRACT FROM AUTHOR]

We consider computation-offloading for Internet-of-things (IoT) applications in multiple-input-multiple-output (MIMO) cloud-radio-access-network (C-RAN). Specifically, the computational tasks of the IoT devices (IoTDs) are offloaded to a MIMO C-RAN, where a MIMO radio resource head (RRH) is connected to a baseband unit (BBU) through a capacity-limited fronthaul link, facilitated by the spatial filtering and uniform scalar quantization. We formulate a computation-offloading optimization problem to minimize the total transmit power of the IoTDs while satisfying the latency requirement of the computational tasks. To obtain a feasible solution for the non-convex problem, firstly the spatial filtering matrix is locally optimized at the MIMO RRH. Subsequently, leveraging the alternating optimization framework for joint optimization on the residual variables at the BBU, the baseband combiner, the optimal resource allocation and the number of quantization bits are obtained through the minimum-mean-squared-error (MMSE) metric, the successive inner convexification method and the line-search method, respectively. As a low-complexity approach, we apply a supervised deep learning (DL) method, which learns from the solutions obtained with our proposed algorithm. In addition, the deep transfer learning is adopted to adjust the neural network in dynamic IoT systems. Numerical results validate the effectiveness of the proposed optimization algorithm and the learning based methods. [ABSTRACT FROM AUTHOR]

Abnormal driving may cause serious danger to both the driver and the public. Existing detectors of abnormal driving behavior are mainly based on shallow models, which require large quantities of labeled data. The acquisition and labelling of abnormal driving data are, however, difficult, labor-intensive and time-consuming. This situation inspires us to rethink the abnormal driving detection problem and to apply deep architecture models. In this study, we establish a novel deep-learning-based model for abnormal driving detection. A stacked sparse autoencoders model is used to learn generic driving behavior features. The model is trained in a greedy layer-wise fashion. As far as the authors know, this is the first time that a deep learning approach is applied using autoencoders as building blocks to represent driving features for abnormal driving detection. In addition, a method for denoising is added to the algorithm to increase the robustness of feature expression. The dropout technology is introduced into the entire training process to avoid overfitting. Experiments carried out on our self-created driving behavior dataset demonstrate that the proposed scheme achieves a superior performance for abnormal driving detection compared to the state-of-the-art. [ABSTRACT FROM AUTHOR]

It is of practical importance to extract and predict taxi hotspots in urban traffic networks. However, the extraction of taxi hotspot is generally influenced by multiple sources of dependence, which has not been well recognized in the existing literature. This study aims to investigate how the integration of clustering models and deep learning approaches can learn and extract the network-wide taxi hotspots in both temporal and spatial dimensions. A density based spatiotemporal clustering algorithm with noise (DBSTCAN) was established to extract the historical taxi hotspots, which changed with time. A conditional generative adversarial network with long short-term memory structure (LSTM-CGAN) model was proposed for taxi hotspot prediction, which is capable of capturing the spatial and temporal variations of taxi hotspots simultaneously. Specifically, the DBSTCAN was applied to process the large-scaled geo-coded taxi pickup data into time-varying historical hotspot information. The proposed LSTM-CGAN model was then trained by the network-wide hotspot data. As illustrated in the numerical tests, it was found that the proposed LSTM-CGAN model provided comparable results with different model layouts and model with 4 LSTM layers in both generator and discriminator performed best. Moreover, four typical prediction approaches were selected to compare with the proposed LSTM-CGAN model. The comparative analyses showed that the proposed LSTM-CGAN model reduced the false positive rate from 5.20% to 0.56% and the false negative rate from 63.50% to 39.93%, and improved the section consistency from 41.78% to 73.43% and the area under the curve from 0.670 to 0.969. The comparison results indicated that the proposed LSTM-CGAN model outperformed all these benchmark methods and demonstrated great potential to enable many shared mobility applications. [ABSTRACT FROM AUTHOR]

The state-of-the-art objects discovery approaches can be categorized into two: deep learning methods based on a convolutional neural network with region proposals, and conventional machine learning methods based on topic models, low-rank matrix factorization, or image processing. The deep learning methods for objects discovery are based on sacrificing computational complexity to achieve precision, and the training time can be long without GPU platforms, whereas the conventional methods usually lack detection accuracy. To effectively address the problems of the training complexity and the detection speed, we present a new objects discovery approach by proposing a two-stage (training and verification) method. In the training stage, a topic model with words co-occurrence prior is proposed on the basis of Latent Dirichlet Allocation (LDA) model, in which the co-occurrence information among the features is sufficiently utilized. In the verification stage, we propose an anchor-box polishing algorithm that fine-tunes the detection results corresponding to the pre-trained topic model from some conventional algorithms with fast detection time. The experiments on various datasets demonstrate that the proposed approach can improve the detection performance in terms of efficiency and computing costs. It is also robust to objects different in colors, lightings, scales, and so on. Interestingly, the proposed method can be combined with many fast but inaccurate detection algorithms, in which it enhances the model’s flexibility. [ABSTRACT FROM AUTHOR]

We present the largest database for visual kinship recognition, Families In the Wild (FIW), with over 13,000 family photos of 1,000 family trees with 4-to-38 members. It took only a small team to build FIW with efficient labeling tools and work-flow. To extend FIW, we further improved upon this process with a novel semi-automatic labeling scheme that used annotated faces and unlabeled text metadata to discover labels, which were then used, along with existing FIW data, for the proposed clustering algorithm that generated label proposals for all newly added data–both processes are shared and compared in depth, showing great savings in time and human input required. Essentially, the clustering algorithm proposed is semi-supervised and uses labeled data to produce more accurate clusters. We statistically compare FIW to related datasets, which unarguably shows enormous gains in overall size and amount of information encapsulated in the labels. We benchmark two tasks, kinship verification and family classification, at scales incomparably larger than ever before. Pre-trained CNN models fine-tuned on FIW outscores other conventional methods and achieved state-of-the art on the renowned KinWild datasets. We also measure human performance on kinship recognition and compare to a fine-tuned CNN. [ABSTRACT FROM AUTHOR]

We propose the Learned Primal-Dual algorithm for tomographic reconstruction. The algorithm accounts for a (possibly non-linear) forward operator in a deep neural network by unrolling a proximal primal-dual optimization method, but where the proximal operators have been replaced with convolutional neural networks. The algorithm is trained end-to-end, working directly from raw measured data and it does not depend on any initial reconstruction such as filtered back-projection (FBP). We compare performance of the proposed method on low dose computed tomography reconstruction against FBP, total variation (TV), and deep learning based post-processing of FBP. For the Shepp-Logan phantom we obtain >6 dB peak signal to noise ratio improvement against all compared methods. For human phantoms the corresponding improvement is 6.6 dB over TV and 2.2 dB over learned post-processing along with a substantial improvement in the structural similarity index. Finally, our algorithm involves only ten forward-back-projection computations, making the method feasible for time critical clinical applications. [ABSTRACT FROM AUTHOR]

Numerous scientific fields rely on elaborate but partly suboptimal data processing pipelines. An example is diffusion magnetic resonance imaging (diffusion MRI), a non-invasive microstructure assessment method with a prominent application in neuroimaging. Advanced diffusion models providing accurate microstructural characterization so far have required long acquisition times and thus have been inapplicable for children and adults who are uncooperative, uncomfortable, or unwell. We show that the long scan time requirements are mainly due to disadvantages of classical data processing. We demonstrate how deep learning, a group of algorithms based on recent advances in the field of artificial neural networks, can be applied to reduce diffusion MRI data processing to a single optimized step. This modification allows obtaining scalar measures from advanced models at twelve-fold reduced scan time and detecting abnormalities without using diffusion models. We set a new state of the art by estimating diffusion kurtosis measures from only 12 data points and neurite orientation dispersion and density measures from only 8 data points. This allows unprecedentedly fast and robust protocols facilitating clinical routine and demonstrates how classical data processing can be streamlined by means of deep learning. [ABSTRACT FROM AUTHOR]