Search

This paper serves as a survey of recent advances in large margin training and its theoretical foundations, mostly for (nonlinear) deep neural networks (DNNs) that are probably the most prominent machine learning models for large-scale data in the community over the past decade. We generalize the formulation of classification margins from classical research to latest DNNs, summarize theoretical connections between the margin, network generalization, and robustness, and introduce recent efforts in enlarging the margins for DNNs comprehensively. Since the viewpoint of different methods is discrepant, we categorize them into groups for ease of comparison and discussion in the paper. Hopefully, our discussions and overview inspire new research work in the community that aim to improve the performance of DNNs, and we also point to directions where the large margin principle can be verified to provide theoretical evidence why certain regularizations for DNNs function well in practice. We managed to shorten the paper such that the crucial spirit of large margin learning and related methods are better emphasized. [ABSTRACT FROM AUTHOR]

Resistive random access memory (ReRAM) is often considered as a strong candidate for storing the weights in non-von Neumann neuromorphic computing systems. This paper studies how nonideal memory characteristics, which include programing error, read fluctuation, and retention, affect the inference accuracy of the analog ReRAM neural networks by incorporating memory characteristics extracted from 1-Mb ReRAM into a simulated inference-only neural network. This paper also shows that the different layer in the network can tolerate different amount of such imperfects. We learned four key points: 1) the conductance range of memory with less relative fluctuation is preferred for designing the weight-conductance mapping; 2) the control of programing error is essential for high inference accuracy; 3) retention-induced conductance drift can be fatal to the neuromorphic system. A compensation scheme is proposed in this paper which can effectively recover the inference accuracy; and 4) for multilayer networks, avoiding weight errors in the front layers can help to maintain the inference accuracy by reducing calculation error which may otherwise accumulate and pass down the networks. The concepts and approaches of this paper can also be applied to evaluate other types of nonvolatile memories for artificial neural networks. [ABSTRACT FROM AUTHOR]

Self-organizing maps (SOM) have been successfully applied in many fields of research. In this paper, we demonstrate the use of SOM-based method for the analysis of Expandable PolyStyrene (EPS) batch process. To this end, a data set of EPS-batch process was used for training a SOM. Reference vectors of the SOM were then classified by K-means algorithm into six clusters, which represent product types of the process. This SOM could also be used for estimating the optimal amounts of the stabilisation agent. The results of a validation data set showed a good agreement between the actual and estimated amounts of the stabilisation agent. Based on this model a Web application was made for test use at the plant. The results indicate that the SOM method can also be efficiently applied to the analysis of the batch process. [ABSTRACT FROM AUTHOR]

In this paper, a model identification method based on a long short-term memory (LSTM) neural network composed of a network structure and training algorithm is used to build a thermal field model that accurately simulates the crystal growth process. The support vector machine (SVM) approach is then adopted to identify model order and lag to determine network input and to improve precision. The thermal field model reflecting the growth process in the Czochralski crystal furnace is simulated. Experimental results and comparative analysis results both suggest that the method proposed by this paper can build an efficient thermal field model which outperforms other methods in terms of precision. [ABSTRACT FROM AUTHOR]

The paper proposes a novel block-wise prediction paradigm based on Convolutional Neural Networks (CNNs) for lossless video coding. A deep neural network model which follows a multi-resolution design is employed for block-wise prediction. Several contributions are proposed to improve neural network training. A first contribution proposes a novel loss function formulation for an efficient network training based on a new approach for patch selection. Another contribution consists in replacing all HEVC-based angular intra-prediction modes with a CNN-based intra-prediction method, where each angular prediction mode is complemented by a CNN-based prediction mode using a specifically trained model. Another contribution consists in an efficient adaptation of the CNN-based intra-prediction residual for lossless video coding. Experimental results on standard test sequences show that the proposed coding system outperforms the HEVC standard with an average bitrate improvement of around 5%. To our knowledge, the paper is the first to replace all the traditional HEVC-based angular intra-prediction modes with an intra-prediction method based on modern Machine Learning techniques for lossless video coding applications. [ABSTRACT FROM AUTHOR]

Distribution system state estimation (DSSE) is a core task for monitoring and control of distribution networks. Widely used algorithms such as Gauss–Newton perform poorly with the limited number of measurements typically available for DSSE, often require many iterations to obtain reasonable results, and sometimes fail to converge. DSSE is a non-convex problem, and working with a limited number of measurements further aggravates the situation, as indeterminacy induces multiple global (in addition to local) minima. Gauss–Newton is also known to be sensitive to initialization. Hence, the situation is far from ideal. It is therefore natural to ask if there is a smart way of initializing Gauss–Newton that will avoid these DSSE-specific pitfalls. This paper proposes using historical or simulation-derived data to train a shallow neural network to “learn to initialize,” that is, map the available measurements to a point in the neighborhood of the true latent states (network voltages), which is used to initialize Gauss–Newton. It is shown that this hybrid machine learning/optimization approach yields superior performance in terms of stability, accuracy, and runtime efficiency, compared to conventional optimization-only approaches. It is also shown that judicious design of the neural network training cost function helps to improve the overall DSSE performance. [ABSTRACT FROM AUTHOR]

With the boom of deep neural networks, blind image quality assessment (BIQA) has achieved great processes. However, the current BIQA metrics are limited when evaluating low-quality images as compared to medium-quality and high-quality images, which restricts their applications in real world problems. In this paper, we first identify that two challenges caused by distribution shift and long-tailed distribution lead to the compromised performance on low-quality images. Then, we propose an intermediary enhancement-based bilateral network with iterative training strategy for solving these two challenges. Drawing on the experience of transitive transfer learning, the proposed metric adaptively introduces enhanced intermediary images to transfer more information to low-quality images for mitigating the distribution shift. Our metric also adopts an iterative training strategy to deal with the long-tailed distribution. This strategy decouples feature extraction and score regression for better representation learning and regressor training. It not only transfers the knowledge learned from the earlier stage to the latter stage, but also makes the model pay more attention to long-tailed low-quality images. We conduct extensive experiments on five authentically distorted image quality datasets. The results show that our metric significantly improves the evaluating performance on low-quality images and delivers state-of-the-art intra-dataset results. During generalization tests, our metric also achieves the best cross-dataset performance. [ABSTRACT FROM AUTHOR]

This paper tackles the problem of training a deep convolutional neural network of both low-bitwidth weights and activations. Optimizing a low-precision network is very challenging due to the non-differentiability of the quantizer, which may result in substantial accuracy loss. To address this, we propose three practical approaches, including (i) progressive quantization; (ii) stochastic precision; and (iii) joint knowledge distillation to improve the network training. First, for progressive quantization, we propose two schemes to progressively find good local minima. Specifically, we propose to first optimize a network with quantized weights and subsequently quantize activations. This is in contrast to the traditional methods which optimize them simultaneously. Furthermore, we propose a second progressive quantization scheme which gradually decreases the bitwidth from high-precision to low-precision during training. Second, to alleviate the excessive training burden due to the multi-round training stages, we further propose a one-stage stochastic precision strategy to randomly sample and quantize sub-networks while keeping other parts in full-precision. Finally, we adopt a novel learning scheme to jointly train a full-precision model alongside the low-precision one. By doing so, the full-precision model provides hints to guide the low-precision model training and significantly improves the performance of the low-precision network. Extensive experiments on various datasets (e.g., CIFAR-100, ImageNet) show the effectiveness of the proposed methods. [ABSTRACT FROM AUTHOR]

In this paper, we introduce a structure-based neural network architecture, namely RC-Struct, for MIMO-OFDM symbol detection. The RC-Struct exploits the temporal structure of the MIMO-OFDM signals through reservoir computing (RC). A binary classifier leverages the repetitive constellation structure in the system to perform multi-class detection. The incorporation of RC allows the RC-Struct to be learned in a purely online fashion with extremely limited pilot symbols in each OFDM subframe. The binary classifier enables the efficient utilization of the precious online training symbols and allows an easy extension to high-order modulations without a substantial increase in complexity. Experiments show that the introduced RC-Struct outperforms both the conventional model-based symbol detection approaches and the state-of-the-art learning-based strategies in terms of bit error rate (BER). The advantages of RC-Struct over existing methods become more significant when rank and link adaptation are adopted. The introduced RC-Struct sheds light on combining communication domain knowledge and learning-based receive processing for 5G/5G-Advanced and Beyond. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks (DCNNs) are currently the method of choice both for generative, as well as for discriminative learning in computer vision and machine learning. The success of DCNNs can be attributed to the careful selection of their building blocks (e.g., residual blocks, rectifiers, sophisticated normalization schemes, to mention but a few). In this paper, we propose $\Pi$ Π -Nets, a new class of function approximators based on polynomial expansions. $\Pi$ Π -Nets are polynomial neural networks, i.e., the output is a high-order polynomial of the input. The unknown parameters, which are naturally represented by high-order tensors, are estimated through a collective tensor factorization with factors sharing. We introduce three tensor decompositions that significantly reduce the number of parameters and show how they can be efficiently implemented by hierarchical neural networks. We empirically demonstrate that $\Pi$ Π -Nets are very expressive and they even produce good results without the use of non-linear activation functions in a large battery of tasks and signals, i.e., images, graphs, and audio. When used in conjunction with activation functions, $\Pi$ Π -Nets produce state-of-the-art results in three challenging tasks, i.e., image generation, face verification and 3D mesh representation learning. The source code is available at https://github.com/grigorisg9gr/polynomial_nets. [ABSTRACT FROM AUTHOR]

A general deep learning (DL) mechanism for a multiple hidden layer feed-forward neural network contains two parts, i.e., 1) an unsupervised greedy layer-wise training and 2) a supervised fine-tuning which is usually an iterative process. Although this mechanism has been demonstrated in many fields to be able to significantly improve the generalization of neural network, there is no clear evidence to show which one of the two parts plays the essential role for the generalization improvement, resulting in an argument within the DL community. Focusing on this argument, this paper proposes a new DL approach to train multilayer feed-forward neural networks. This approach uses restricted Boltzmann machine (RBM) as the layer-wise training and uses the generalized inverse of a matrix as the supervised fine-tuning. Different from the general deep training mechanism like back-propagation (BP), the proposed approach does not need to iteratively tune the weights, and therefore, has many advantages such as quick training, better generalization, and high understandability, etc. Experimentally, the proposed approach demonstrates an excellent performance in comparison with BP-based DL and the traditional training method for multilayer random weight neural networks. To a great extent, this paper demonstrates that the supervised part plays a more important role than the unsupervised part in DL, which provides some new viewpoints for exploring the essence of DL. [ABSTRACT FROM AUTHOR]

Despite encouraging results have been achieved in human pose estimation in recent years, it remains challenging problems. When the background is similar to the human body parts, and there are small persons with low-resolution in the image, the performance may degrade dramatically. This paper addresses problems in background-inference and small-person pose estimation. To achieve this, a novel pose estimation algorithm is proposed on the basis of person semantic segmentation deep neural network. Different from most previous methods with a single pose estimation model, we generate mixture models with pose estimation and semantic segmentation. We introduce novel generative adversarial model and auxiliary model to realize the semantic segmentation network, which can handle the confusion of the similar regions in the background. In addition, to address the problem of the scale differences between big and small persons’ keypoints, we add additional position and channel attention modules to the first two stages of OpenPose. We conduct extensive experiments on COCO and VOC datasets. And we compare the proposed method with the most popular state-of-the-art human pose estimation and semantic segmentation frameworks, including MultiPoseNet, Deterton2 and DeepLab V3. Our experimental results show that the proposed method is more accurate than the state-of-the-art algorithms and performs effectively in tackling the complex situations. [ABSTRACT FROM AUTHOR]

Image retargeting techniques adjust images into different sizes and have attracted much attention recently. Objective quality assessment (OQA) of image retargeting results is often desired to automatically select the best results. Existing OQA methods train a model using some benchmarks (e.g., RetargetMe), in which subjective scores evaluated by users are provided. Observing that it is challenging even for human subjects to give consistent scores for retargeting results of different source images (diff-source-results), in this paper we propose a learning-based OQA method that trains a General Regression Neural Network (GRNN) model based on relative scores—which preserve the ranking—of retargeting results of the same source image (same-source-results). In particular, we develop a novel training scheme with provable convergence that learns a common base scalar for same-source-results. With this source specific offset, our computed scores not only preserve the ranking of subjective scores for same-source-results, but also provide a reference to compare the diff-source-results. We train and evaluate our GRNN model using human preference data collected in RetargetMe. We further introduce a subjective benchmark to evaluate the generalizability of different OQA methods. Experimental results demonstrate that our method outperforms ten representative OQA methods in ranking prediction and has better generalizability to different datasets. [ABSTRACT FROM AUTHOR]

The weights of a multilayer perceptron (MLP) may be altered by multiplicative and/or additive noises if it is implemented in hardware. Moreover, if an MLP is implemented using analog circuits, it is prone to stuck-at 0 faults, i.e., link failures. In this paper, we have proposed a methodology for making an MLP robust with respect to link failures, multiplicative noise, and additive noise. This is achieved by penalizing the system error with three regularizing terms. To train the system we use a weighted sum of the following four terms: 1) mean squared error (MSE); 2) ${l^{2}}$ norm of the weight vector; 3) sum of squares of the first order derivatives of MSE with respect to weights; and 4) sum of squares of the second order derivatives of MSE with respect to weights. The proposed approach has been tested on ten regression and ten classification tasks with link failure, multiplicative noise, and additive noise scenarios. Our experimental results demonstrate the effectiveness of the proposed regularization to achieve robust training of an MLP. [ABSTRACT FROM AUTHOR]

The accurate reconstruction of areas obscured by clouds is among the most challenging topics for the remote sensing community since a significant percentage of images archived throughout the world are affected by cloud covers which make them not fully exploitable. The purpose of this paper is to propose new methods to recover missing data in multispectral images due to the presence of clouds by relying on a formulation based on an autoencoder (AE) neural network. We suppose that clouds are opaque and their detection is performed by dedicated algorithms. The AE in our methods aims at modeling the relationship between a given cloud-free image (source image) and a cloud-contaminated image (target image). In particular, two strategies are developed: the first one performs the mapping at a pixel level while the second one at a patch level to take profit from spatial contextual information. Moreover, in order to fix the problem of the hidden layer size, a new solution combining the minimum descriptive length criterion and a Pareto-like selection procedure is introduced. The results of experiments conducted on three different data sets are reported and discussed together with a comparison with reference techniques. [ABSTRACT FROM AUTHOR]

Neural networks (NNs) are attractive for nonlinear impairment compensation applications in communication systems, such as optical fiber nonlinearity, nonlinearity of driving amplifiers, and nonlinearity of semiconductor optical amplifiers. Without prior knowledge of the transmission link or the hardware characteristics, optimal parameters are completely constructed from a data-driven approach by exploring training datasets, once the NN structure is given. On the other hand, due to computational power and energy consumption, especially in high-speed communication systems, the computational complexity of the optimized NN needs to be confined to the hardware, such as FPGA or ASIC without sacrificing its performance improvement. In this paper, two approaches are presented to accommodate the NN-based algorithms for high-speed communication systems. The first approach is to reduce computational complexity of the NN-based nonlinearity compensation algorithms on the basis of weight pruning (WP). WP can significantly reduce the computational complexity, especially because the nonlinear compensation task studied here results in a sparse NN. The authors have studied an enhanced approach of WP by imposing an additional restriction on the selection of non-zero weights on each hidden layer. The second approach is to implement NNs onto a silicon-photonic integrated platform, enabling power efficiency to be further improved without sacrificing the high-speed operation. [ABSTRACT FROM AUTHOR]

Residual networks (ResNets) have introduced a milestone for the deep learning community due to their outstanding performance in diverse applications. They enable efficient training of increasingly deep networks, reducing the training difficulty and error. The main intuition behind them is that, instead of mapping the input information, they are mapping a residual part of it. Since the original work, a lot of extensions have been proposed to improve information mapping. In this paper, a novel extension of the residual block is proposed inspired by linear dynamical systems (LDSs), called LDS-ResNet. Specifically, a new module is presented that improves mapping of residual information by transforming it in a hidden state and then mapping it back to the desired feature space using convolutional layers. The proposed module is utilized to construct multi-branch residual blocks for convolutional neural networks. An exploration of possible architectural choices is presented and evaluated. Experimental results show that LDS-ResNet outperforms the original ResNet in image classification and object detection tasks on public datasets such as CIFAR-10/100, ImageNet, VOC, and MOT2017. Moreover, its performance boost is complementary to other extensions of the original network such as pre-activation and bottleneck, as well as stochastic training and Squeeze-Excitation. [ABSTRACT FROM AUTHOR]

Video super-resolution (VSR) has become one of the most critical problems in video processing. In the deep learning literature, recent works have shown the benefits of using adversarial-based and perceptual losses to improve the performance on various image restoration tasks; however, these have yet to be applied for video super-resolution. In this paper, we propose a generative adversarial network (GAN)-based formulation for VSR. We introduce a new generator network optimized for the VSR problem, named VSRResNet, along with new discriminator architecture to properly guide VSRResNet during the GAN training. We further enhance our VSR GAN formulation with two regularizers, a distance loss in feature-space and pixel-space, to obtain our final VSRResFeatGAN model. We show that pre-training our generator with the mean-squared-error loss only quantitatively surpasses the current state-of-the-art VSR models. Finally, we employ the PercepDist metric to compare the state-of-the-art VSR models. We show that this metric more accurately evaluates the perceptual quality of SR solutions obtained from neural networks, compared with the commonly used PSNR/SSIM metrics. Finally, we show that our proposed model, the VSRResFeatGAN model, outperforms the current state-of-the-art SR models, both quantitatively and qualitatively. [ABSTRACT FROM AUTHOR]

The demand and relevance of efficient analog fault diagnosis methods for modern RF and microwave-integrated circuits increase with the growing need and complexity of analog and mixed-signal circuitry. The well-established digital fault diagnosis methods are insufficient for analog circuitry due to the intrinsic complexity in analog faults and their corresponding identification process. In this paper, we present an artificial neural network (ANN) modeling approach to efficiently emulate the injection of analog faults in RF circuits. The resulting metamodel is used for fault identification by applying an optimization-based process using a constrained parameter extraction formulation. A generalized neural modeling formulation to include auxiliary measurements in the circuit is proposed. This generalized formulation significantly increases the uniqueness of the faults identification process. The proposed methodology is illustrated by two faulty analog circuits: a CMOS RF voltage amplifier and a reconfigurable bandpass microstrip filter. [ABSTRACT FROM AUTHOR]

Deep learning (DL) aims at learning the meaningful representations. A meaningful representation gives rise to significant performance improvement of associated machine learning (ML) tasks by replacing the raw data as the input. However, optimal architecture design and model parameter estimation in DL algorithms are widely considered to be intractable. Evolutionary algorithms are much preferable for complex and nonconvex problems due to its inherent characteristics of gradient-free and insensitivity to the local optimal. In this paper, we propose a computationally economical algorithm for evolving unsupervised deep neural networks to efficiently learn meaningful representations, which is very suitable in the current big data era where sufficient labeled data for training is often expensive to acquire. In the proposed algorithm, finding an appropriate architecture and the initialized parameter values for an ML task at hand is modeled by one computational efficient gene encoding approach, which is employed to effectively model the task with a large number of parameters. In addition, a local search strategy is incorporated to facilitate the exploitation search for further improving the performance. Furthermore, a small proportion labeled data is utilized during evolution search to guarantee the learned representations to be meaningful. The performance of the proposed algorithm has been thoroughly investigated over classification tasks. Specifically, error classification rate on MNIST with 1.15% is reached by the proposed algorithm consistently, which is considered a very promising result against state-of-the-art unsupervised DL algorithms. [ABSTRACT FROM AUTHOR]

In this paper, we aim to build a comprehensive face detection system which provides a one-stop solution to various practical challenges for face detection in realistic scenarios, e.g., detecting faces from multiple-views, faces with occlusions, exaggerated expressions or blurred faces. Moreover, we introduce an automatic data harvest algorithm to effectively improve the generalization performance of the system even when collecting training faces containing various challenging patterns is difficult. In particular, we introduce three critical components to build the system, i.e., a recently widely used deep convolutional neural network (CNN), a novel blur-aware bi-channel network architecture, and a new self-learning mechanism capable of exploiting video contexts continuously. The aforementioned challenges except for detecting blurred faces can potentially be addressed by the CNN component owing its robustness to local deformation of target faces. The more challenging problem of detecting blurred faces is addressed by the bi-channel architecture component which processes blurred and clear faces adaptively. In addition, to address the difficulties in improving the generalization performance of the learning-based face detection system, we introduce a video-context-based self-learning mechanism into the system, which enables the system to continuously enhance its performance by harvesting faces with challenging training patterns automatically. To exploit video context, the detector is applied to massive unlabeled videos, and challenging faces are captured based on temporal inference. These recaptured faces, generally corresponding to one or multiple challenges mentioned above, are fed into the detection system to further improve its performance. Extensive experiments with the proposed detection system provide new state-of-the-art performance on FDDB data set, PASCAL face data set, AFW data set, and WIDER Face data set. [ABSTRACT FROM AUTHOR]

The main challenges of visual object tracking arise from the arbitrary appearance of the objects that need to be tracked. Most existing algorithms try to solve this problem by training a new model to regenerate or classify each tracked object. As a result, the model needs to be initialized and retrained for each new object. In this paper, we propose to track different objects in an object-independent approach with a novel two-flow convolutional neural network (YCNN). The YCNN takes two inputs (one is an object image patch, the other is a larger searching image patch), then outputs a response map which predicts how likely and where the object would appear in the search patch. Unlike the object-specific approaches, the YCNN is actually trained to measure the similarity between the two image patches. Thus, this model will not be limited to any specific object. Furthermore, the network is end-to-end trained to extract both shallow and deep dedicated convolutional features for visual tracking. And once properly trained, the YCNN can be used to track all kinds of objects without further training and updating. As a result, our algorithm is able to run at a very high speed of 45 frames-per-second. The effectiveness of the proposed algorithm can also be proved by the experiments on two popular data sets: OTB-100 and VOT-2014. [ABSTRACT FROM AUTHOR]

Classification of 3D objects – the selection of a category in which each object belongs – is of great interest in the field of machine learning. Numerous researchers use deep neural networks to address this problem, altering the network architecture and representation of the 3D shape used as an input. To investigate the effectiveness of their approaches, we conduct an extensive survey of existing methods and identify common ideas by which we categorize them into a taxonomy. Second, we evaluate 11 selected classification networks on two 3D object datasets, extending the evaluation to a larger dataset on which most of the selected approaches have not been tested yet. For this, we provide a framework for converting shapes from common 3D mesh formats into formats native to each network, and for training and evaluating different classification approaches on this data. Despite being partially unable to reach the accuracies reported in the original papers, we compare the relative performance of the approaches as well as their performance when changing datasets as the only variable to provide valuable insights into performance on different kinds of data. We make our code available to simplify running training experiments with multiple neural networks with different prerequisites. [ABSTRACT FROM AUTHOR]

Federated Learning (FL) enables the multiple participating devices to collaboratively contribute to a global neural network model while keeping the training data locally. Unlike the centralized training setting, the non-IID, imbalanced (statistical heterogeneity) and distribution shifted training data of FL is distributed in the federated network, which will increase the divergences between the local models and the global model, further degrading performance. In this paper, we propose a flexible clustered federated learning (CFL) framework named FlexCFL, in which we 1) group the training of clients based on the similarities between the clients’ optimization directions for lower training divergence; 2) implement an efficient newcomer device cold start mechanism for framework scalability and practicality; 3) flexibly migrate clients to meet the challenge of client-level data distribution shift. FlexCFL can achieve improvements by dividing joint optimization into groups of sub-optimization and can strike a balance between accuracy and communication efficiency in the distribution shift environment. The convergence and complexity are analyzed to demonstrate the efficiency of FlexCFL. We also evaluate FlexCFL on several open datasets and made comparisons with related CFL frameworks. The results show that FlexCFL can significantly improve absolute test accuracy by $+10.6\%$ + 10. 6 % on FEMNIST compared with FedAvg, $+3.5\%$ + 3. 5 % on FashionMNIST compared with FedProx, $+8.4\%$ + 8. 4 % on MNIST compared with FeSEM, $+4.7\%$ + 4. 7 % on Sentiment140 compare with IFCA. The experiment results show that FlexCFL is also communication efficient in the distribution shift environment. [ABSTRACT FROM AUTHOR]

In this paper, a new ensemble forecasting model for short-term load forecasting (STLF) is proposed based on extreme learning machine (ELM). Four important improvements are used to support the ELM for increased forecasting performance. First, a novel wavelet-based ensemble scheme is carried out to generate the individual ELM-based forecasters. Second, a hybrid learning algorithm blending ELM and the Levenberg–Marquardt method is proposed to improve the learning accuracy of neural networks. Third, a feature selection method based on the conditional mutual information is developed to select a compact set of input variables for the forecasting model. Fourth, to realize an accurate ensemble forecast, partial least squares regression is utilized as a combining approach to aggregate the individual forecasts. Numerical testing shows that proposed method can obtain better forecasting results in comparison with other standard and state-of-the-art methods. [ABSTRACT FROM AUTHOR]

The high voltage insulator requires continuous monitoring and inspection to prevent failures and emergencies. Manual inspections are costly as it requires covering a large geographical area where insulators are often subjected to harsh weather conditions. Automatic detection of insulators from aerial images is the first step towards performing real-time classification of insulator conditions using Unmanned Aerial Vehicle (UAV). In this paper, we provide a cost-effective solution for detecting insulators under the conditions of an uncluttered background, varied object resolution and illumination conditions using You Only Look Once (YOLO) deep learning neural network model from aerial images. We apply data augmentation to avoid overfitting with a training set size of 56000 image samples. It is demonstrated experimentally that this method can accurately locate insulator on UAV based real-time image data. The detected insulator images are then successfully subjected to insulator surface condition assessment for the presence of ice, snow and water using different classifiers. [ABSTRACT FROM AUTHOR]

This paper presents a new technique for artificial neural network (ANN) inverse modeling and applications to microwave filters. In inverse modeling of a microwave component, the inputs to the model are electrical parameters such as $S$ -parameters, and the outputs of the model are geometrical or physical parameters. Since the analytical formula of the inverse input–output relationship does not exist, the ANN becomes a logical choice, because it can be trained to learn from the data in inverse modeling. The main challenge of inverse modeling is the nonuniqueness problem. This problem in the ANN inverse modeling is that different training samples with the same or very similar input values have quite different (contradictory) output values (multivalued solutions). In this paper, we propose a multivalued neural network inverse modeling technique to associate a single set of electrical parameters with multiple sets of geometrical or physical parameters. One set of geometrical or physical parameters is called one value of our proposed inverse model. Our proposed multivalued neural network is structured to accommodate multiple values for the model output. We also propose a new training error function to focus on matching each training sample using only one value of our proposed inverse model, while other values are free and can be trained to match other contradictory samples. In this way, our proposed multivalued neural network can learn all the training data by automatically redirecting contradictory information into different values of the proposed inverse model. Therefore, our proposed technique can solve the nonuniqueness problem in a simpler and more automated way compared with the existing ANN inverse modeling techniques. This technique is illustrated by inverse modeling and parameter extraction of four microwave filter examples. [ABSTRACT FROM AUTHOR]

A residual networks family with hundreds or even thousands of layers dominates major image recognition tasks, but building a network by simply stacking residual blocks inevitably limits its optimization ability. This paper proposes a novel residual network architecture, residual networks of residual networks (RoR), to dig the optimization ability of residual networks. RoR substitutes optimizing residual mapping of residual mapping for optimizing original residual mapping. In particular, RoR adds levelwise shortcut connections upon original residual networks to promote the learning capability of residual networks. More importantly, RoR can be applied to various kinds of residual networks (ResNets, Pre-ResNets, and WRN) and significantly boost their performance. Our experiments demonstrate the effectiveness and versatility of RoR, where it achieves the best performance in all residual-network-like structures. Our RoR-3-WRN58-4 + SD models achieve new state-of-the-art results on CIFAR-10, CIFAR-100, and SVHN, with the test errors of 3.77%, 19.73%, and 1.59%, respectively. RoR-3 models also achieve state-of-the-art results compared with ResNets on the ImageNet data set. [ABSTRACT FROM AUTHOR]

In this paper, we introduce a new neural network (NN) structure, multi-mode reservoir computing (Multi-Mode RC). It inherits the dynamic mechanism of RC and processes the forward path and loss optimization of the NN using tensor as the underlying data format. Multi-Mode RC exhibits less complexity compared to conventional RC structures (e.g. single-mode RC), and offers comparable generalization performance to its single-mode counterpart. Furthermore, we introduce an alternating least square-based learning algorithm as well as the associated theoretical analysis for Multi-Mode RC. The result can be utilized to guide the configuration of NN parameters to sufficiently circumvent over-fitting issues. As a key application, we consider the symbol detection task in multiple-input-multiple-output (MIMO) orthogonal-frequency-division-multiplexing (OFDM) systems with massive MIMO employed at the base stations (BSs). Thanks to the tensor structure of massive MIMO-OFDM signals, our online learning-based symbol detection method generalizes well in terms of bit error rate even using a limited online training set. Evaluation results suggest that the Multi-Mode RC-based learning framework can efficiently and effectively combat practical constraints of wireless systems (i.e. channel state information (CSI) errors and hardware non-linearity) to enable robust and adaptive communications over the air. [ABSTRACT FROM AUTHOR]

Error disagreement-based active learning (AL) selects the data that maximally update the error of a classification hypothesis. However, poor human supervision (e.g., few labels, improper classifier parameters) may weaken or clutter this update; moreover, the computational cost of performing a greedy search to estimate the errors using a deep neural network is intolerable. In this paper, a novel disagreement coefficient based on distribution, not error, provides a tighter bound on label complexity, which further guarantees its generalization in hyperbolic space. The focal points derived from the squared Lorentzian distance, present more effective hyperbolic representations on aspherical distribution from geometry, replacing the typical euclidean, kernelized, and Poincaré centroids. Experiments on different deep AL tasks show that, the focal representation adopted in a tree-likeliness splitting, significantly performs better than typical baselines of centroid representations and error disagreement, and state-of-the-art neural network architectures-based AL, dramatically accelerating the learning process. [ABSTRACT FROM AUTHOR]

With substantial amount of time, resources and human (team) efforts invested to explore and develop successful deep neural networks (DNN), there emerges an urgent need to protect these inventions from being illegally copied, redistributed, or abused without respecting the intellectual properties of legitimate owners. Following recent progresses along this line, we investigate a number of watermark-based DNN ownership verification methods in the face of ambiguity attacks, which aim to cast doubts on the ownership verification by forging counterfeit watermarks. It is shown that ambiguity attacks pose serious threats to existing DNN watermarking methods. As remedies to the above-mentioned loophole, this paper proposes novel passport-based DNN ownership verification schemes which are both robust to network modifications and resilient to ambiguity attacks. The gist of embedding digital passports is to design and train DNN models in a way such that, the DNN inference performance of an original task will be significantly deteriorated due to forged passports. In other words, genuine passports are not only verified by looking for the predefined signatures, but also reasserted by the unyielding DNN model inference performances. Extensive experimental results justify the effectiveness of the proposed passport-based DNN ownership verification schemes. Code is available at https://github.com/kamwoh/DeepIPR. [ABSTRACT FROM AUTHOR]

Reliably quantifying the confidence of deep neural classifiers is a challenging yet fundamental requirement for deploying such models in safety-critical applications. In this paper, we introduce a novel target criterion for model confidence, namely the true class probability (TCP). We show that TCP offers better properties for confidence estimation than standard maximum class probability (MCP). Since the true class is by essence unknown at test time, we propose to learn TCP criterion from data with an auxiliary model, introducing a specific learning scheme adapted to this context. We evaluate our approach on the task of failure prediction and of self-training with pseudo-labels for domain adaptation, which both necessitate effective confidence estimates. Extensive experiments are conducted for validating the relevance of the proposed approach in each task. We study various network architectures and experiment with small and large datasets for image classification and semantic segmentation. In every tested benchmark, our approach outperforms strong baselines. [ABSTRACT FROM AUTHOR]

Most studies about deep learning are based on neural network models, where many layers of parameterized nonlinear differentiable modules are trained by backpropagation. Recently, it has been shown that deep learning can also be realized by non-differentiable modules without backpropagation training called deep forest. We identify that deep forest has high time costs and memory requirements—this has inhibited its use on large-scale datasets. In this paper, we propose a simple and effective approach with three main strategies for efficient learning of deep forest. First, it substantially reduces the number of instances that needs to be processed through redirecting instances having high predictive confidence straight to the final level for prediction, by-passing all the intermediate levels. Second, many non-informative features are screened out, and only the informative ones are used for learning at each level. Third, an unsupervised feature transformation procedure is proposed to replace the supervised multi-grained scanning procedure. Our theoretical analysis supports the proposed approach in varying the model complexity from low to high as the number of levels increases in deep forest. Experiments show that our approach achieves highly competitive predictive performance with reduced time cost and memory requirement by one to two orders of magnitude. [ABSTRACT FROM AUTHOR]

Remarkable achievements have been obtained by deep neural networks in the last several years. However, the breakthrough in neural networks accuracy is always accompanied by explosive growth of computation and parameters, which leads to a severe limitation of model deployment. In this paper, we propose a novel knowledge distillation technique named self-distillation to address this problem. Self-distillation attaches several attention modules and shallow classifiers at different depths of neural networks and distills knowledge from the deepest classifier to the shallower classifiers. Different from the conventional knowledge distillation methods where the knowledge of the teacher model is transferred to another student model, self-distillation can be considered as knowledge transfer in the same model - from the deeper layers to the shallow layers. Moreover, the additional classifiers in self-distillation allow the neural network to work in a dynamic manner, which leads to a much higher acceleration. Experiments demonstrate that self-distillation has consistent and significant effectiveness on various neural networks and datasets. On average, 3.49 and 2.32 percent accuracy boost are observed on CIFAR100 and ImageNet. Besides, experiments show that self-distillation can be combined with other model compression methods, including knowledge distillation, pruning and lightweight model design. [ABSTRACT FROM AUTHOR]

Stochastic optimization algorithms have been popular for training deep neural networks. Recently, there emerges a new approach of learning-based optimizer, which has achieved promising performance for training neural networks. However, these black-box learning-based optimizers do not fully take advantage of the experience in human-designed optimizers and heavily rely on learning from meta-training tasks, therefore have limited generalization ability. In this paper, we propose a novel optimizer, dubbed as Variational HyperAdam, which is based on a parametric generalized Adam algorithm, i.e., HyperAdam, in a variational framework. With Variational HyperAdam as optimizer for training neural network, the parameter update vector of the neural network at each training step is considered as random variable, whose approximate posterior distribution given the training data and current network parameter vector is predicted by Variational HyperAdam. The parameter update vector for network training is sampled from this approximate posterior distribution. Specifically, in Variational HyperAdam, we design a learnable generalized Adam algorithm for estimating expectation, paired with a VarBlock for estimating the variance of the approximate posterior distribution of parameter update vector. The Variational HyperAdam is learned in a meta-learning approach with meta-training loss derived by variational inference. Experiments verify that the learned Variational HyperAdam achieved state-of-the-art network training performance for various types of networks on different datasets, such as multilayer perceptron, CNN, LSTM and ResNet. [ABSTRACT FROM AUTHOR]

Scanner sensitivity is often critical in high-resolution Positron Emission Tomography (PET) dedicated to molecular imaging. In neighboring pixelated detectors with individual readout, sensitivity decreases because of multiple coincidences produced by Compton scattering. Correct analysis of those coincidences would enable a substantial sensitivity increase. However, including scattering byproducts in the image often lead to image quality degradation because of inaccurate Line-of-Response (LOR) assessment. In such scanners, to support high count rates, multiple coincidences are usually discarded when image degradation is not acceptable, or blindly accepted for a low computational burden. This paper presents a new, real-time capable method that includes Inter-Crystal Scatter (ICS) triple coincidences in the image without significant quality degradation. The method computes the LOR using a neural network fed by preprocessed raw data. As a proof of principle, this paper analyzes the simplest ICS scenario, triple coincidences where one photoelectric 511-keV event coincides with two more whose energy sum is also 511 keV. The paper visits the algorithm structure, presents Monte Carlo assessment with the LabPET model, and displays images reconstructed from real data. With an energy window of 360-660 keV and a singles energy threshold of 125 keV, the inclusion of triple coincidences yielded a sensitivity increase of 54%, a resolution degradation similar to that of other sensitivity-increasing methods, and only a slight contrast degradation for real LabPET data, with potential for numerous further improvements. [ABSTRACT FROM PUBLISHER]

Despite the tremendous success of deep neural networks in various learning problems, it has been observed that adding intentionally designed adversarial perturbations to inputs of these architectures leads to erroneous classification with high confidence in the prediction. In this work, we show that adversarial examples can be generated using a generic approach that relies on the perturbation analysis of learning algorithms. Formulated as a convex program, the proposed approach retrieves many current adversarial attacks as special cases. It is used to propose novel attacks against learning algorithms for classification and regression tasks under various new constraints with closed-form solutions in many instances. In particular, we derive new attacks against classification algorithms which are shown to be top-performing on various architectures. Although classification tasks have been the main focus of adversarial attacks, we use the proposed approach to generate adversarial perturbations for various regression tasks. Designed for single pixel and single subset attacks, these attacks are applied to autoencoding, image colorization and real-time object detection tasks, showing that adversarial perturbations can degrade equally gravely the output of regression tasks. 1 In the spirit of encouraging reproducible research, the implementations used in this paper have been made available at: github.com/ebalda/adversarialconvex. [ABSTRACT FROM AUTHOR]

In optical transport networks, failure localization is usually triggered as a response to alarms and significant anomalous behaviors. However, the recent evolution of network control and management leveraging software-defined networking (SDN) and streaming-based telemetry opens up new possibilities for automated methods that can localize even subtle anomalies, the so-called soft failures. This paper reports the experimental demonstration of a machine-learning-based soft-failure localization framework in a small-scale laboratory setup. The SDN telemetry setup includes ONOS-controlled transponders using NETCONF and an optical line system (OLS) providing telemetry via an OLS domain controller. A shallow artificial neural network (ANN) accomplishes ML-based failure localization with principal component analysis to reduce non-essential information. The ANN is trained by combining field data and synthetic data generated in a digital network twin. The field data trains the ANN to tolerate statistical variations in the network telemetry without failures, while the synthetic data generates artificial single-failure scenarios. We show that the soft-failure localization scheme successfully pinpoints the faulty element in all single failures generated in transponders, fibers, and amplifiers. We also demonstrate the system’s ability to deal with double-failure scenarios. [ABSTRACT FROM AUTHOR]

Semantic segmentation methods based on deep neural networks have achieved great success in recent years. However, training such deep neural networks relies heavily on a large number of images with accurate pixel-level labels, which requires a huge amount of human effort, especially for large-scale remote sensing images. In this paper, we propose a point-based weakly supervised learning framework called the deep bilateral filtering network (DBFNet) for the semantic segmentation of remote sensing images. Compared with pixel-level labels, point annotations are usually sparse and cannot reveal the complete structure of the objects; they also lack boundary information, thus resulting in incomplete prediction within the object and the loss of object boundaries. To address these problems, we incorporate the bilateral filtering technique into deeply learned representations in two respects. First, since a target object contains smooth regions that always belong to the same category, we perform deep bilateral filtering (DBF) to filter the deep features by a nonlinear combination of nearby feature values, which encourages the nearby and similar features to become closer, thus achieving a consistent prediction in the smooth region. In addition, the DBF can distinguish the boundary by enlarging the distance between the features on different sides of the edge, thus preserving the boundary information well. Experimental results on two widely used datasets, the ISPRS 2-D semantic labeling Potsdam and Vaihingen datasets, demonstrate that our proposed DBFNet can achieve a highly competitive performance compared with state-of-the-art fully-supervised methods. Code is available at https://github.com/Luffy03/DBFNet. [ABSTRACT FROM AUTHOR]

Representation learning has been proven to play an important role in the unprecedented success of machine learning models in numerous tasks, such as machine translation, face recognition and recommendation. The majority of existing representation learning approaches often require a large number of consistent and noise-free labels. However, due to various reasons such as budget constraints and privacy concerns, labels are very limited in many real-world scenarios. Directly applying standard representation learning approaches on small labeled data sets will easily run into over-fitting problems and lead to sub-optimal solutions. Even worse, in some domains such as education, the limited labels are usually annotated by multiple workers with diverse expertise, which yields noises and inconsistency in such crowdsourcing settings. In this paper, we propose a novel framework which aims to learn effective representations from limited data with crowdsourced labels. Specifically, we design a grouping based deep neural network to learn embeddings from a limited number of training samples and present a Bayesian confidence estimator to capture the inconsistency among crowdsourced labels. Furthermore, to expedite the training process, we develop a hard example selection procedure to adaptively pick up training examples that are misclassified by the model. Extensive experiments conducted on three real-world data sets demonstrate the superiority of our framework on learning representations from limited data with crowdsourced labels, comparing with various state-of-the-art baselines. In addition, we provide a comprehensive analysis on each of the main components of our proposed framework and also introduce the promising results it achieved in our real production to fully understand the proposed framework. To encourage reproducible results, we make our code available online at https://github.com/tal-ai/RECLE. [ABSTRACT FROM AUTHOR]

In time-varying fading channels, channel coefficients are estimated using pilot symbols that are transmitted every coherence interval. For channels with high Doppler spread, the rapid channel variations over time will require considerable bandwidth for pilot transmission, leading to poor throughput. In this paper, we propose a novel receiver architecture using deep recurrent neural networks (RNNs) that learns the channel variations and thereby reduces the number of pilot symbols required for channel estimation. Specifically, we design and train an RNN to learn the correlation in the time-varying channel and predict the channel coefficients into the future with good accuracy over a wide range of Dopplers and signal-to-noise ratios (SNR). The proposed training methodology enables accurate channel prediction through the use of techniques such as teacher-force training, early-stop, and reduction of learning rate on plateau. Also, the robustness of prediction for different Dopplers and SNRs is achieved by adapting the number of predictions into the future based on the Doppler and SNR. Numerical results show that good bit error performance is achieved by the proposed receiver in time-varying fading channels. We also propose a data decision driven receiver architecture using RNNs that further reduces the pilot overhead while maintaining good bit error performance. [ABSTRACT FROM AUTHOR]

In-band full duplex communication requires precise self-interference cancellation (SIC) to successfully decode the received desired signal. The existing neural network (NN) based SIC scheme uses offline trained NN to estimate the non-linear component of received self-interference (SI) over a static SI channel without additional NN training. For the time-varying SI channel, the NN-aided SIC method needs to retrain the NN during in-band full duplex communication to adapt time-varying channels. However, NN training is not fast enough to be done during full duplex communication. Thus, the SIC performance degrades. In this paper, we propose a digital SIC scheme using channel robust NN which takes estimates of linear SI channel coefficients as an input. It was found that the NN could learn the static part of the non-linear behavior of the SI channel well. Our solution can adapt the time-varying SI channel with only the channel coefficient estimation of linear components and a pre-trained NN. The proposed scheme can successfully reduce the SI to the noise floor for both time-invariant and time-varying SI channels. [ABSTRACT FROM AUTHOR]

In general, an image attribute is a human-nameable visual property that has a semantic connotation. Appropriate modeling of the intrinsic contextual correlations among attributes plays a fundamental role in attribute detection. In this paper, we consider image attribute detection from the perspective of regularized deep learning. In particular, we propose a regularized deep belief network (rDBN) to perform the image attribute detection task, which is composed of two parts: 1) a detection DBN (dDBN) that models the joint distribution of images and their corresponding attributes, which acts as an attribute detector and 2) a contextual restricted Boltzmann machine that explicitly models the correlations among attributes acting as a regularizer that restraints the output detection result given by the dDBN to meet the contextual prior of attributes. Furthermore, we propose an efficient fine-tuning scheme that can further optimize the performance of the dDBN by backpropagation. Experimental results show that the proposed rDBN obtains improvements over the state-of-the-art methods for attribute detection on the benchmark data sets. [ABSTRACT FROM PUBLISHER]

Gramophone records were the main recording medium for more than seven decades and regained widespread popularity over the past several years. Being an analog storage medium, gramophone records are subject to distortions caused by scratches, dust particles, degradation, and other means of improper handling. The observed noise often leads to an unpleasant listening experience and requires a filtering process to remove the unwanted disruptions and improve the audio quality. This paper proposes a novel approach that employs various feed forward time delay artificial neural networks to detect and reconstruct noise in musical sound waves. A set of 800 songs from eight different genres were used to validate the performance of the neural networks. The performance was analyzed according to the outlier detection and interpolation accuracy, the computational time and the tradeoff between the accuracy and the time. The empirical results of both detection and reconstruction neural networks were compared to a number of other algorithms, including various statistical measurements, duplication approaches, trigonometric processes, polynomials, and time series models. It was found that the neural networks’ outlier detection accuracy was slightly lower than some of the other noise identification algorithms, but achieved a more efficient tradeoff by detecting most of the noise in real time. The reconstruction process favored neural networks with an increase in the interpolation accuracy compared to other widely used time series models. It was also found that certain genres such as classical, country, and jazz music were interpolated more accurately. Volatile signals, such as electronic, metal, and pop music were more challenging to reconstruct and were substantially better interpolated using neural networks than the other examined algorithms. [ABSTRACT FROM PUBLISHER]