Search

The deep learning era endows immense opportunities for ubiquitous robotic applications by leveraging big data generated from widespread sensors and ever-growing computing capability. While the growing demands for natural human-robot interaction (HRI) as well as concerns for energy efficiency, real-time performance, and data security motive novel solutions. In this paper, we present a brain-inspired spiking neural network (SNN) based Human-Robot Auditory Interface, namely HuRAI. The HuRAI integrates the voice activity detection, speaker localization and voice command recognition systems into a unified framework that can be implemented on the emerging low-power neuromorphic computing (NC) devices. Our experimental results demonstrate superior modeling capabilities of SNNs, achieving accurate and rapid prediction for each task. Moreover, the energy efficiency analysis reveals a compelling prospect, with up to three orders of magnitude energy savings, over the equivalent artificial neural networks that running on the state-of-the-art Nvidia graphics processing unit (GPU). Therefore, integrating the algorithmic power of large-scale SNN models and the energy efficiency of NC devices offers an attractive solution for real-time, low-power robotic applications. [ABSTRACT FROM AUTHOR]

In this paper, Harris' Hawks weight optimization-guided artificial neural learning-based quicker session key coordination for Industrial Internet-of-Things (IIoT) to enhance the security of Critical Energy Infrastructure (CEI) is proposed. Transportation, telecommunications, healthcare, finance, and defense are all being revolutionized by the energy industry's digitization. CEI is widely dispersed, resulting in complex cyber-physical networks that require constant monitoring and quick recovery to avoid cyberattacks. Substantial efforts were made in this regard to tackle the key exchange problem in IIoT devices, the majority of which have depended on traditional approaches. Existing solutions fail to adequately resolve the security and privacy issues that IIoT systems face. This study proposes a Triple Layer Vector-Valued Neural Network (TLVVNN) to cope with the problem. However, research into optimizing the value of neural weights for quicker neural synchronization is rare. In this case, Harris' Hawks is used to optimizing the neural network's weight vector for quicker coordination. The coordinated weight becomes the session key once this process is accomplished. This technique has several advantages, including (1) Generation of session key via mutual neural synchronization over the public channel. (2) It enables Harris' Hawks-based neural weight vector optimization for faster neural synchronization across public channels. (3) Vector inputs and weights are taken into consideration for TLVVNN networks. (4) The internal structure of the TLVVNN is complex by three hidden layers. As a result, the attacker might have a lot of difficulties determining the internal architecture. (5) Several pairs of variable-length session keys are generated by TLVVNN. (6) It prevents impersonation, geometric, brute force, and majority assaults. Tests to validate the performance of the proposed methodology are carried out, and the results show that the proposed methodology outperforms similar approaches already in use. [ABSTRACT FROM AUTHOR]

• Analysis of effect of neuronal leak in spiking neural network models. • Leak enhances model robustness by cutting off some high frequency components from input. • Leaky model also decreases sparsity of computation. • Analysis relating leak to input frequency component showing lowpass filtering of leak Spiking Neural Networks (SNNs) are being explored to emulate the astounding capabilities of human brain that can learn to perform robust and efficient computations with noisy spikes. A variety of spiking neuron models have been proposed to resemble biological neuronal functionalities. The simplest and most commonly used among these SNNs are leaky-integrate-and-fire (LIF), which contain a leak path in their membrane potential and integrate-and-fire (IF), where the leakage path is absent. While the LIF models have been argued as more bio-plausible, a comparative analysis between models with and without leak from a purely computational point of view demands attention, which we try to address in this paper. Our results reveal that LIF model provides improved robustness and better generalization compared to IF. Frequency domain analysis demonstrates that leak aids in eliminating high-frequency components from the input, thus enhancing noise-robustness of SNNs. Additionally, we compare the sparsity of computation between these models. In general, for the same input, the LIF model would be expected to achieve higher sparsity compared to IF due to the layer-wise decay of spikes caused by membrane potential leak with time. However, contrary to this expectation, we observe that leak decreases the sparsity of computation. Therefore, there exists a trade-off between robustness and energy-efficiency in SNNs which can be optimized through suitable choice of amount of leak in the models. [ABSTRACT FROM AUTHOR]

This paper is devoted to developing a biological-based algorithm to simulate the control of a human arm by means of a Spiking Neural Network (SNN) with a pre-set structure similar to those found in reflex arcs. The replication of the behavior of biological systems is a primary step to understand how the brain and the neural networks that comprise it work. The study of these systems began with the analysis of animals with low neural complexity as well as with small control neural networks such as those present in the reflex acts of the human body. However, the study of the brain, due to its intricate structure, still presents many unknowns and challenges to be solved. Meanwhile, one way to understand how biological systems work is to emulate their behavior through computer simulations. Artificial neural networks (ANNs) offered the opportunity to replicate neural structures to understand and reproduce their behavior and performance. Many types of ANNs are based on the use of activation functions. However, these ANNs are simplified models that do not replicate the behavior of complex biological neural systems accurately. For this reason, spike-based models have been developed to reproduce real biological systems more faithfully. This work proposes simulating the motor behavior of the central nervous system to control the position of an arm. To this end, a spiking neural network has been developed to emulate motion control using a fixed structure that reproduces reflex arcs. A channel-based synapse model and a control scheme based on the equilibrium point hypothesis have been proposed to improve biological similarity of the controller. Furthermore, the developed controller has learning capabilities thanks to the reproduction of the synapse plasticity process that takes place in real biological systems through a supervised Spike-Timing Dependent Plasticity (STDP) learning strategy. The performance of the proposed approach has been demonstrated by simulating the control of the movement of an arm using Hill's Muscle Model. Results suggest that the proposed control algorithm resembles the response of biological systems adequately in terms of speed and temporal response. [ABSTRACT FROM AUTHOR]

Recently, region proposal network (RPN) has been combined with the Siamese network for tracking and shown excellent accuracy and high efficiency. However, the low correlation between the classification score and localization accuracy in tracking has weakened the performance of the tracking model. In this paper, we propose an IoU-guided Siamese RPN (SiamIG) to address this problem. Specifically, SiamIG predicts the intersection-over-union (IoU) with the ground truth for each regressed anchor by using an IoU predictor. Then the predicted IoU is multiplied with the classification score to compute the final score, which is more suitable for localization accuracy. Furthermore, each anchor in the regression branch is assigned an IoU-based weighting factor such that the tracking accuracy can be further improved. Specifically, anchor boxes with high IoU are given more attention because of the IoU-based weighting factor, which helps the model pay more attention to an anchor box with a high IoU. Experiments demonstrate that the proposed method runs at over 200 fps and achieves up to 2.4%(AUC), 3%(EAO) and 2.3%(AUC) improvements over the baseline on the OTB-100, VOT-2016 and UAV123 datasets respectively. [ABSTRACT FROM AUTHOR]

• A time–frequency deep metric learning model is proposed for MTS classification. • A consistency regularizer is designed to capture the correlations among levels. • An effective optimization strategy is developed to train the model hierarchically. • Results on 18 MTS datasets show that our method outperforms other methods. Multivariate time series (MTS) data exist in various fields of studies and MTS classification is an important research topic in the machine learning community. Researchers have proposed many MTS classification models over the years and the distance-based methods along with nearest neighbor classifier achieve good performance. However, the current methods mainly focus on defining distance metric on time-domain of MTS and ignore frequency information. Besides, these methods usually define the same linear distance metric for different datasets, which is not suitable for capturing the nonlinear relationship of MTS and degrades the discriminative power of the distance metric. In this paper, we propose a time–frequency deep metric learning (TFDM) approach for MTS classification. The multilevel discrete wavelet decomposition is first adopted to decompose an MTS into a group of sub-MTS so as to extract multilevel time–frequency representations. Then, a deep convolutional neural network is developed for each level to learn level-specific nonlinear features and a metric learning layer is added on the top of the network to learn the semantic similarity of MTS. Moreover, a cross-level consistency regularization term is designed to encourage the distance metrics of different levels to be consistent for capturing the correlations among different levels. Finally, we use 1-nearest neighbor to classify MTS according to the learned distance metrics. Extensive experiments on 18 benchmark datasets show the effectiveness of our approach. [ABSTRACT FROM AUTHOR]

This paper focuses on an adaptive learning and control problem for a class of discrete-time nonlinear uncertain systems operating under multiple environments. A novel intelligent learning control framework is proposed by using a combination of offline and online learning methods. Specifically, in the offline learning mode, a deterministic learning (DL) based adaptive dynamics learning approach is first proposed to achieve locally-accurate identification of associated nonlinear uncertain system dynamics under each anticipated individual environment, and the learned knowledge is obtained and stored in a set of constant radial basis function neural network models. Then, with the learned knowledge, an online adaptive learning control scheme is further developed, which consists of: (i) an online adaptive learning control mechanism composed of multiple experience-based controllers and a DL-based adaptive learning controller, aiming to provide desired control performance for the plant operating under each individual environment; and (ii) a learning-based recognition mechanism composed of multiple recognition estimators and a DL-based identifier, aiming to recognize the active environment and schedule appropriate control strategies in real time. To guarantee the system stability during environment transition, a robust quasi-sliding-mode controller is further developed and embedded in the overall controller architecture. With this new intelligent adaptive learning control framework, the overall system is capable of adapting not only to any anticipated (pre-defined) environment by re-utilizing the knowledge obtained from both offline and online learning, but also to unanticipated (new) environments by actively acquiring new knowledge online. Simulation studies are conducted to verify the effectiveness and advantages of this new framework. [ABSTRACT FROM AUTHOR]

[Display omitted] We think that there are mainly two issues to be considered in few-shot learning, one is how to learn a generalized representation and the other is how to alleviate over-fitting. For the first issue, in contrast to current few-shot learning models which directly take VGG or ResNet as an embedding network, we propose to design a novel embedding network to improve the generalization ability on unseen classes. For the second issue, we propose to deal with few-shot learning problem in a semi-supervised setting. The unlabeled set is used to compensate the lack of few-shot training samples. As illustrated in Graphical abstract figure, the samples in the unlabeled set are from not only known classes but also distracter classes, which is more realistic. Instead of taking pair-wise distance measures to select unlabeled samples, we propose a contextual similarity considering both local pair-wise distance and the global task manifold to pick and label known unlabeled samples. Then, we train an episodic linear classifier based on the combination of the support set and the picked known unlabeled samples. The parameters of the proposed multi-level second-order attention network will be updated by minimizing the loss of the query set. • To improve the generalization ability of the embedding network to unseen classes, we design a novel multi-level second-order attention network to capture the correlations of the intermediate layers. • To pick and label the quasi-known unlabeled instances, we propose a novel and effective contextual similarity measure jointly considering the pair-wise similarity and global similarity. Extensive experiments on four datasets (Omniglot, miniImageNet, tieredImageNet, and CUB-200-2011) show the proposed method is superior in performance. In this paper, we tackle the few-shot learning problem in a semi-supervised setting where a limited number of labeled data-points and a number of low-cost unlabeled samples are assumed to be available. In particular, some of the unlabeled samples share the same label space with the support set, referring to as known samples, while some of them are from distracter classes, referring to as unknown ones. The keys are how to learn a powerful representation and how to pick and label unlabeled known instances to construct discriminative classifiers. We address both issues by learning multi-level second-order attention representation followed by a contextual similarity. We first develop a novel trainable multi-level second-order attention network(MSAN) to adaptively learn more powerful feature representation by using second-order feature statistics. Our proposed MSAN is able to better represent the samples while the parameter is not increased. Then we introduce a contextual measure that considers not only the pair-wise relationship but also the task-specific condition, to calculate the similarity between unlabeled samples and each support class, thus to label and pick the quasi-known samples. The hypothesis is that the known unlabeled sample should not only be strongly similar to one particular class, but also be significantly dissimilar to other classes. With the combination of support set and pseudo-labeled set, we train an episodic linear classifier for each episode and the parameters of multi-level second-order attention network are updated by minimizing the loss of the query set. Extensive experiments on four popular benchmarks (Omniglot, miniImageNet, tieredImageNet, and CUB-200-2011) demonstrate that our simple yet effective approach can achieve competitive accuracy compared to the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Neural networks can be used as a data-driven model for system identification. But the probability properties of the training data are not included. The Bayesian approach can model the input and output probability distribution, but it cannot give points estimation. In this paper, we propose a special neural model, which combines the neural model and Bayesian inference. The probability distributions of input and output are obtained by the Bayesian approach. This statistical model changes the neural network's structure and improve the accuracy of the neural modeling. We also propose a neural network training method using Bayesian inference. The approach capabilities are analyzed. We use three examples to compare our method with the other black box methods. The results show that this new model is much better, when there are large noises, and the dynamics of the system is complex. [ABSTRACT FROM AUTHOR]

• Developing a new recursive convolutional layer (RCL) able to stop the iteration when the hidden state stabilize. • First network to have different depth for different batch of samples. • A comprehensive study on the best locations and numbers of RCL in a multilayer network called C-FRPN. To overcome problems with the design of large networks, particularly with respect to the depth of the network, this paper presents a new model of convolutional neural networks (CNN) which features fully recursive convolutional layers (RCLs). An RCL is a generalization of the classic one-stage feedforward convolutional layer (CL) to fully direct feedback connections between the outputs of the CL and its inputs. A traditional deep CNN consisting of many CLs, can then be generalized to include some CLs, and some RCLs in the intermediate stages. We call the corresponding network a Convolutional Neural Network with Fully Recursive Perceptron Network (C-FRPN). Through an analysis of results obtained from applications of the C-FRPN to three benchmark image classification datasets: CIFAR-10, SVHN, ISIC, it is found that (i) in general, the performance of a C-FRPN, even with only one RCL, is better than the performance of the corresponding deep CNN with all CLs, under the constraint of having the same number of unknown parameters; (ii) the performance of the C-FRPN varies with respect to (a) where the RCLs are located, and (b) the number of RCLs in the C-FRPN; and, (iii) the effectiveness of the RCLs depends on the size of the training dataset. The results suggest that: (a) it is advisable to use RCLs particularly when training very large sets of data, (b) it is best to prioritize placement of RCLs close to the input layer of the C-FRPN, and (c) it is advisable to increase the number of RCLs as long as the training dataset can sustain without overfitting being observed. [ABSTRACT FROM AUTHOR]

This paper proposes a change detection algorithm based on Fully Convolutional Siamese Networks for optical aerial images, which is trained by using an improved Contrastive Loss-Focal Contrastive Loss (FCL). The proposed framework equipped with contrastive loss can extract features directly from image pairs and measure changes by using a distance metric. In other words, this method encourages reducing intra-class variance and enlarging inter-class difference, so that the binarized change map can be obtained by a simple threshold. In change detection task, a critical problem is how to overcome example imbalance (i.e. unchanged examples are much more than changed examples). To address this challenge, a novel focal contrastive loss is proposed to further improve the performance of the model. FCL can reduce the impact of example imbalance and make the model focus learning on hard examples. Extensive experiments demonstrate that the proposed approach is more abstract as well as robust. Compared with other baseline methods, the presented method achieves better results on SAZDA, TISZADOB, CDD, and WHU-CD data set. It achieves state-of-the-art performance in terms of the F1 measure. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks have shown great potential in image recognition tasks. However, the fact that the mechanism of deep learning is difficult to explain hinders its development. It involves a large amount of parameter learning, which results in high computational complexity. Moreover, deep convolutional neural networks are often limited by overfitting in regimes in which the number of training samples is limited. Conversely, kernel learning methods have a clear mathematical theory, fewer parameters, and can contend with small sample sizes; however, they are not able to handle high-dimensional data, e.g., images. It is important to achieve a performance and complexity trade-off in complicated tasks. In this paper, we propose a novel scalable deep convolutional random kernel learning in Gaussian process architecture called SDCRKL-GP, which is characterized by excellent performance and low complexity. First, we successfully incorporated the deep convolutional architecture into kernel learning by implementing the random Fourier feature transform for Gaussian processes, which can effectively capture hierarchical and local image-level features. This approach enabled the kernel method to effectively handle image processing problems. Second, we optimized the parameters of deep convolutional filters and Gaussian kernels by stochastic variational inference. Then, we derived the lower variational bound of the marginal likelihood. Finally, we explored the model architecture design space selection method to determine the appropriate network architecture for different datasets. The design space consists of the number of layers, the channels per layer, and so on. Different design space selections improved the scalability of the SDCRKL-GP architecture. We evaluated SDCRKL-GP on the MNIST, FMNIST, CIFAR10, and CALTECH4 benchmark datasets. Taking MNIST as an example, the error rate of classification is 0.60%, and the number of parameters, number of computations and memory access cost of the architecture are 19.088k, 0.984M, and 1.057M, respectively. The experimental results verified that the proposed SDCRKL-GP method outperforms several state-of-the-art algorithms in both accuracy and speed in image recognition tasks. The code is available at https://github.com/w-tingting/deep-rff-pytorch. [ABSTRACT FROM AUTHOR]

Interpersonal relation plays an essential role to gain understandings on how people interact with each other. In computer vision, interpersonal relations provide vital information to interpret people's behaviors. However, the existing research has either omitted the interaction information between subjects or the structural information in the images. In this paper, we propose a new architecture to reason interpersonal relations based on higher-order graph networks and multi-scale features. First, we extract features of the whole images, the facial features, and the union region of face pairs. Apart from the pixel-wise features, we also consider the positional features of face-to-face pairs and the spatial scene cues. Higher-order Graph Neural Networks (GNNs) were employed to map out the interpersonal relations based on the feature extracted. Experimental results show that the proposed Higher-order Graph Neural Networks with multi-scale features can effectively recognize the social relations in images with over 5% improvement in absolute balanced accuracy compared with the state-of-the-art work. [ABSTRACT FROM AUTHOR]

Fast and accurate structural dynamics analysis is important for structural design and damage assessment. Structural dynamics analysis leveraging machine learning techniques has become a popular research focus in recent years. Although the basic neural network provides an alternative approach for structural dynamics analysis, the lack of physics law inside the neural network limits the model accuracy and fidelity. In this paper, a new family of the energy-conservation neural network is introduced, which respects the physical laws. The neural network is explored from a fundamental single-degree-of-freedom system to a complicated multiple-degrees-of-freedom system. The damping force and external forces are also considered step by step. To improve the parallelization of the algorithm, the derivatives of the structural states are parameterized with the novel energy-conservation neural network instead of specifying the discrete sequence of structural states. The proposed model uses the system energy as the last layer of the neural network and leverages the underlying automatic differentiation graph to incorporate the system energy naturally, which ultimately improves the accuracy and long-term stability of structural dynamics response calculation under an earthquake impact. The trade-off between computation accuracy and speed is discussed. As a case study, a 3-story building earthquake simulation is conducted with realistic earthquake records. [ABSTRACT FROM AUTHOR]

Knowledge distillation (KD) builds a lightweight Student Model (SM) and trains it to approximate a large Teacher Model (TM) by exploring knowledge learned by the TM, which shows effectiveness to train lightweight CNN models. However, training a small SM to achieve better performance remains a challenging problem. Recent researches on human learning behaviors show that both the knowledge from teachers and the knowledge learning processes of teachers are significant for students. Inspired by this characteristic, in this paper, we propose a new perspective, called Progressive Mimic Learning (PML), to train lightweight CNN models by mimicking the learning trajectory of the TM. In order to obtain a more powerful SM, the useful hints in the learning process of the TM are explored. To start with, the TM learning process is divided into multiple stages, and the last state of the TM in each stage is recorded as a landmark. The learning trajectory of the TM is composed of these landmarks. Then, a landmark loss is defined to constrain the SM to progressively mimic the learning process of the TM, by employing landmarks in the learning trajectory as a training hint of the SM. Several experiments are conducted on four benchmark data sets, CIFAR-10, CIFAR-100, Fashion-MNIST, and ImageNet-10, to investigate the performance of the PML. The results show that the PML can make SMs generate more accurate predictions than SMs trained by its counterparts. [ABSTRACT FROM AUTHOR]

Sketch recognition remains a significant challenge due to the limited training data and the substantial intra-class variance of freehand sketches for the same object. Conventional methods for this task often rely on the availability of the temporal order of sketch strokes, additional cues acquired from different modalities and supervised augmentation of sketch datasets with real images, which also limit the applicability and feasibility of these methods in real scenarios. In this paper, we propose a novel sketch-specific data augmentation (SSDA) method that leverages the quantity and quality of the sketches automatically. From the aspect of quantity, we introduce a Bezier pivot based deformation (BPD) strategy to enrich the training data. Towards quality improvement, we present a mean stroke reconstruction (MSR) approach to generate a set of novel types of sketches with smaller intra-class variances. Both of these solutions are unrestricted from any multi-source data and temporal cues of sketches. Furthermore, we show that some recent deep convolutional neural network models that are trained on generic classes of real images can be better choices than most of the elaborate architectures that are designed explicitly for sketch recognition. As SSDA can be integrated with any convolutional neural networks, it has a distinct advantage over the existing methods. Our extensive experimental evaluations demonstrate that the proposed method achieves the state-of-the-art results (84.27%) on the TU-Berlin dataset, outperforming the human performance by a remarkable 11.17% increase. Finally, more experiments show the practical value of our approach for the task of sketch-based image retrieval. [ABSTRACT FROM AUTHOR]

While supervised learning approaches show great vitality and effectiveness in video object segmentation, most of them require large amounts of annotations which are expensive and time-consuming. Recently, self-supervised learning has attracted great attention by benefiting from unlabeled video sequences. However, current patch-based self-supervised video object segmentation methods only discriminate the patch from the entire image without distinguishing the object of interest from meaningless backgrounds or even occlusion. These disturbances deteriorate the extracted features and hinder the robustness of tracking when applied to real-world video sequences. In this paper, we propose a novel model named Tracker With Integration-Augmented Attention (TWIAA) to achieve both label-free and prominent performance. Specifically, we integrate both spatial and channel dimensions by introducing a feature spatial enhancement module and a two-stream channel module. With the combination of the two modules, the network can focus on exploring the discriminative object and suppressing the irrelevant part to improve the tracking robustness. Moreover, unlike other methods that calculate features separately on the search branch and template branch, the two designed modules coupled with the Siamese network compute the respective features of the search branch and the template branch jointly to augment the interdependence of the two branches. Such interdependence is injected into both spatial and channel dimensions. So that our approach establishes richer and more discriminative associations to identify the object more accurately. In addition, our method takes full advantage of cycle-consistency information in consecutive frames, which uses coherence as the learning signal to acquire object-oriented relationships. Extensive experiments and ablation studies are conducted on large VOS benchmarks, including DAVIS-2017, YouTube-VOS-2018, and YouTube-VOS-2019. The results verify that our proposed framework has both strong feature representation and competitive performance compared with supervised and self-supervised models. [ABSTRACT FROM AUTHOR]

Object detection based on deep learning typically relies on a large number of training data, which may be very labor-consuming to prepare. In this paper, we attempt to tackle the problem by addressing the One-Shot Object Detection (OSOD) task. Given a novel image denoted as the query image whose category label is not included in the training data, OSOD aims to detect objects of the same class in a complex scene denoted as the target image. The performance of recent OSOD methods is much weaker than general object detection. We find that one of the reasons behind this limited performance is that more false positives (i.e., false detections) are generated. Therefore, we argue that it is important to reduce the number of false positives generated in OSOD task to improve performance. To this end, we present a Focus On Classification One-Shot Object Detection (FOC OSOD) network. Specifically, we design the network from two perspectives: (1) how to obtain the effective similarity feature between the query image and target image; (2) how to classify the similarity feature effectively. To solve the above two challenges, firstly, we propose a Classification Feature Deformation-and-Attention (CFDA) module to obtain the high-quality query feature and target feature, so we can further generate effective similarity feature between them. Secondly, we present a Split Iterative Head (SIH) to improve the ability to classify the similarity feature. Extensive experiments on two public datasets (i.e., PASCAL VOC and COCO) demonstrate that the proposed framework achieves superior performance which outperforms other state-of-the-art methods with a considerable margin. [ABSTRACT FROM AUTHOR]

Weakly Supervised Object Localization (WSOL) is a technique for obtaining the object location from attention maps of the classification network, without using bounding box annotations. Existing WSOL approaches lack the modeling of the correlations between different regions of the target object. Hence they can only locate some discriminative attentions, which are small and sparse. Besides, they introduce too many background attentions when mining more object parts. In this paper, we propose a novel Adaptive Attention Augmentor (A3) to adaptively augment the target object attentions on class attention maps. It can supplement object attentions by discovering the semantic correspondence between different regions and dynamically suppress background attentions through the proposed Focal Dice loss. Extensive experiments demonstrate the effectiveness of our approach. On the ILSVRC dataset, A3 achieves a new state-of-the-art localization performance. On the fine-grained datasets including CUB-200–2011 and Cars-196, it also achieves very competitive results. [ABSTRACT FROM AUTHOR]

The biology-inspired intelligent computing system for the neuromorphic hardware implementation is useful in high-speed parallel information processing. However, the traditional Von Neumann computer architecture and the unsatisfactory signal transmission approach have jointly limited the overall performance of the specific hardware implementation. In this paper, a compact extreme learning machine (ELM) architecture synthesized with the spintronic memristor-based synaptic circuit, the biasing circuit, and the activation function circuit is presented. Notably, due to the threshold characteristic of the memristive device, the synaptic circuit has a bimodal behavior. Namely, it is capable to provide the constant and adjustable network weights between the adjacent layers in the ELM. Furthermore, two major limitations (process variations and sneak path issue) are taken into account for the detailed robustness analysis of the whole network. Finally, the entire scheme is verified with case studies in single image super-resolution (SR) reconstruction. [ABSTRACT FROM AUTHOR]

Person re-identification (Re-ID) is of important capability for artificial intelligence and human–computer interaction. The main challenge of person Re-ID lies in limited data to precisely capture a wide range of appearance variations over multiple viewpoints. Furthermore, compared to the Re-ID between the single person, the person groups contain information about the relationship between pedestrians that can potentially help identify certain identities. The Re-ID combining groups and individuals remain to be a promising task under rare study. In this paper, we propose a harmonious attention network for person re-identification, in which we jointly consider the complementarity between person groups and individuals. Concretely, first we propose a two-stream attentive network (TSAN) to respectively learn the information from the person groups and individuals. TSAN consists of a spatial–temporal fusion network for the group Re-ID, as well as a deep network for the traditionally individual person Re-ID. To jointly consider the contributions of the groups and individuals, then we propose a novel re-ranking algorithm (GIRK) based on the learned features to associate the group and individual information. We also propose a new group Re-ID dataset DukeGroupVid to evaluate the performance of our approach. Comprehensive experimental results on the proposed dataset and other Re-ID datasets demonstrate the effectiveness of our model. [ABSTRACT FROM AUTHOR]

It has been proved that deep neural networks are facing a new threat called backdoor attacks, where the adversary can inject backdoors into the neural network model through poisoning the training dataset. When the input containing some special pattern called the backdoor trigger, the model with backdoor will carry out malicious task such as misclassification specified by adversaries. In text classification systems, backdoors inserted in the models can cause spam or malicious speech to escape detection. Previous work mainly focused on the defense of backdoor attacks in computer vision, little attention has been paid to defense method for RNN backdoor attacks regarding text classification. In this paper, through analyzing the changes in inner LSTM neurons, we proposed a defense method called Backdoor Keyword Identification (BKI) to mitigate backdoor attacks which the adversary performs against LSTM-based text classification by data poisoning. This method can identify and exclude poisoning samples crafted to insert backdoor into the model from training data without a verified and trusted dataset. We evaluate our method on four different text classification datset: IMDB, DBpedia ontology, 20 newsgroups and Reuters-21578 dataset. It all achieves good performance regardless of the trigger sentences. [ABSTRACT FROM AUTHOR]

Artificial neural networks (ANNs) are known to be sensitive to the initial setting of parameters and the network architecture, such as the number of hidden nodes in multilayer perceptron (MLP). In this paper, we focus on a network structure which can help to find the proper number of hidden nodes in MLP. In this structure, so called Multilayer Perceptrons with Hidden Multipliers (MLPHM), each of the hidden nodes is associated with a tunable "gate" multiplier. With a specific regularization term, each gate tends to be opened or closed completely at the end of the training, and finally a pruned network is obtained. To study the fault tolerance and to improve the generalization of MLPHM, a noise-injected training scheme is proposed, with both multiplicative noise and additive noise taken into consideration. The objective functions and convergence theorems of the noise-injected training algorithms are obtained, and the latter have been verified by simulations. Applications to several UCI datasets have demonstrated that the proposed algorithms have efficient pruning ability and superior generalization ability. [ABSTRACT FROM AUTHOR]

DeepLab—one of the most critical deep neural network models for image segmentation—has achieved the most advanced performance in the field of image semantic understanding. However, the rich parameters and calculation of deep convolutional neural networks (DCNNs) demand further research on neuro-inspired computing chips specifically designed for hardware acceleration and AI applications. This motivates memristive solutions—the memristor integrates a range of merits, such as fast speed and nanoscale device, to provide an ideal implementation for building low-power neural computing chips and ultra-high-density non-volatile memory. Here, this paper proposes a memristive DeepLab (MDeepLab) system with software-hardware co-design, in which atrous convolution is used to enlarge the receptive field of neurons, the measure to avoid the gridding effect is adopted. Also, the weights are stored by leveraging the unit of a single crossbar array and the constant-term circuit, which significantly reduces the number of memristor devices, doubles the density, and nearly halves the power consumption conventional crossbar array. Finally, the effectiveness of the proposed scheme is verified by a series of experimental simulations and result analysis, showing the superiority of MDeepLab. This study is expected to provide a new solution for low-power consumption and real-time image processing of edge devices. [ABSTRACT FROM AUTHOR]

Dropout is an important training method for deep neural networks, because it can help avoid over-fitting. Traditional dropout methods and many extended dropout methods, omit some of the neurons' activation values according to the probabilities. These methods calculate the activation probability of neurons using the designed formula, without providing a plausible explanation of the calculation method. This paper proposes an adaptive sparse dropout (AS-Dropout) method for neural network training. The algorithm maps the neurons' activation values in a layer to a relative linear range of a sigmoid function, determines the ratio of active neurons by a probability calculation process, and drops most of neurons according to the probabilities. The probability calculation depends on the activation values of the neurons. The selection of active neurons is according to the probabilities. Therefore, AS-Dropout learns both the certainty and uncertainty in deep neural networks. Additionally, since only a small number of neurons are active, AS-Dropout increases the sparsity of the network. We applied AS-Dropout in different neural network structures. When evaluated on MNIST, COIL-100, and Caltech-101 datasets, the experimental results demonstrated that, overall, AS-Dropout substantially outperformed the traditional dropout and some improved dropout methods. [ABSTRACT FROM AUTHOR]

At present, multi-oriented text detection methods based on deep neural network have achieved promising performances on various benchmarks. Nevertheless, there are still some difficulties for arbitrary shape text detection, especially for a simple and proper representation of arbitrary shape text instances. In this paper, a pixel-based text detector is proposed to facilitate the representation and prediction of text instances with arbitrary shapes in a simple manner. Firstly, to alleviate the influence of the target vertex sorting and achieve the direct regression of arbitrary shape text instances, the starting-point-independent coordinates regression loss is proposed. Furthermore, to predict more accurate text instances, the text instance accuracy loss is proposed as an assistant task to refine the predicted coordinates under the guidance of IoU. To evaluate the effectiveness of our detector, extensive experiments have been carried on public benchmarks which contain arbitrary shape text instances and multi-oriented text instances. We obtain 84.8% of F-measure on Total-Text benchmark. The results show that our method can reach state-of-the-art performance. [ABSTRACT FROM AUTHOR]

Data visualization plays an important role in data analysis by displaying data to observers in an interpretable way. Visualizing multidimensional data requires projecting the data into a low-dimensional space that is visible to humans. In this paper, we propose a neural network model that can generate such projections while preserving the topology relationships within data points, which is named Visible Self Organizing Incremental Neural Network (V-SOINN). V-SOINN is able to construct a topology preserving visible network automatically and classify visible nodes to different classes in the low-dimensional space. The thought of topology preserving visualization stems from Self-Organizing Map (SOM). Compared to SOM, the main advantage of V-SOINN is that it does not need prior decision of network structure, including the number of nodes and grid in the output layer. V-SOINN can show the density distribution of datasets by using the activation counts of datasets. V-SOINN is able to depict the number of classes in the low-dimensional space as well. We perform experiments on artificial and real-world datasets, and V-SOINN outperforms PCA, MDS, t-SNE, Neural Gas and SOM on the datasets. Experiments show that V-SOINN can preserve the topology and V-SOINN can produce the correct classification result when the number of samples is small. [ABSTRACT FROM AUTHOR]

The existed methods for electroencephalograph (EEG) emotion recognition always train the models based on all the EEG samples indistinguishably. However, some of the source (training) samples may lead to a negative influence because they are significant dissimilar with the target (test) samples. So it is necessary to give more attention to the EEG samples with strong transferability rather than forcefully training a classification model by all the samples. Furthermore, for an EEG sample, from the aspect of neuroscience, not all the brain regions of an EEG sample contain emotional information that can transferred to the test data effectively. Even some brain region data will make strong negative effect for learning the emotional classification model. Considering these two issues, in this paper, we propose a transferable attention neural network (TANN) for EEG emotion recognition, which learns the emotional discriminative information by highlighting the transferable EEG brain regions data and samples adaptively through local and global attention mechanism. This can be implemented by measuring the outputs of multiple brain-region-level discriminators and one single sample-level discriminator. Extensive experiments on EEG emotion recognition demonstrate that the proposed TANN is superior to those state-of-the-art methods. [ABSTRACT FROM AUTHOR]

This paper investigates the adaptive neural network optimal output feedback control design problem for nonlinear continuous-time systems with actuator saturation. The system dynamics and states of the controlled system are unknown. A neural network state observer is constructed to estimate the system states. This paper uses two neural networks, one is used to construct the neural network state observer, the other (critic neural network) is used to approximate the cost functions, which comprise an observer-critic architecture. In this architecture, the critic neural network weights are tuned based on both the current data and the previous data, thus the conditions of the persistent excitation in the previous literatures are relaxed. By utilizing adaptive dynamic programming approach, a new observer-based optimal control scheme is developed. It is proved that the proposed adaptive neural network output feedback optimal control scheme can ensure that the whole closed-loop system is stable. Moreover, the estimate errors of the critic neural network weights are asymptotically stable. A simulation example is given to validate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

This paper investigates the problem of the exponential synchronization of complex dynamical networks with time-varying inner coupling via event-triggered communication. The network topology is assumed to have a spanning tree. A sufficient condition is derived to guarantee the exponential synchronization by employing the special Lyapunov stability analysis method, which by combining the difference and differential of the Lyapunov function rather than the single difference or differential. The main advantage of this paper is to avoid continuous communication between network nodes, which can decrease the number of information updates, reduce the network congestion and avoid the waste of network resources. Moreover, the Zeno behavior is excluded as well by the strictly positive sampling intervals. Finally, A simulation example is given to show the effectiveness of the proposed exponential synchronization criteria. [ABSTRACT FROM AUTHOR]

Multi-task learning employs a shared representation of knowledge for learning several instances of the same problem. Multi-step time series problem is one of the most challenging problems for machine learning methods. The performance of a prediction model face challenges for higher prediction horizons due to the accumulation of errors. Cooperative coevolution employs in a divide and conquer approach for training neural networks and has been very promising for single step ahead time series prediction. Recently, co-evolutionary multi-task learning has been proposed for dynamic time series prediction. In this paper, we adapt co-evolutionary multi-task learning for multi-step prediction where predictive recurrence is developed to feature knowledge from previous states for future prediction horizon. The goal of the paper is to present a network architecture with predictive recurrence which is capable of multi-step prediction through a form of multi-task learning. We employ cooperative neuro-evolution and an evolutionary algorithm as baselines for comparison. The results show that the proposed method provides the best generalization performance in most cases. Comparison of results with the literature has shown to be promising which motivates further application of the approach for related real-world problems. [ABSTRACT FROM AUTHOR]

In this paper, we study the cooperative output regulation problem for discrete-time linear time-delay multi-agent systems subject to jointly connected switching networks. Our distributed control law is composed of a purely decentralized control law and a distributed switched observer. In contrast to the same problem with static networks, the closed-loop system of this paper is a switched system, the stability analysis of the closed-loop system cannot be conducted by the existing frequency domain technique. Therefore, we need to establish two lemmas to lay the foundation for solving the problem and then present the solution to the problem by a distributed dynamic output feedback control law. An example is used to illustrate our approach. [ABSTRACT FROM AUTHOR]

To compact the architecture of extreme learning machine (ELM), two incremental learning algorithms are proposed in this paper. The previous incremental learning algorithms for ELM recruit hidden nodes randomly, which is equivalent to implementing a random selection from a candidate set of infinite size. Hence, it is impossible to recruit good hidden nodes, and thus it usually requires more hidden nodes than traditional neural networks to achieve matched performance. To improve the quality of the hidden nodes recruited, an incremental learning algorithm for ELM is presented based on Gram--Schmidt process (GSI-ELM), which recruits the best hidden node from a random subset of fixed size via defining an evaluating criterion at each learning step. However, the “nesting effect” exists in the GSI-ELM, that is to say, the hidden nodes once recruited by GSI-ELM can not be later discarded. To treat this “nesting problem”, the improved GSI-ELM (IGSI-ELM) is generated with an elimination mechanism. At each learning step IGSI-ELM eliminates the worst hidden node from the already-recruited group if it is not the newly-recruited one. Finally, to verify the efficacy and feasibility of the proposed algorithms, i.e. GSI-ELM and IGSI-ELM, in this paper, experiments on regression and classification benchmark data sets are investigated. [ABSTRACT FROM AUTHOR]

This paper presents an artificial neural network to solve the quadratic zero-one programming problems under linear constraints. In this paper, by using the connection between integer and nonlinear programming, the quadratic zero-one programming problem is transformed into the quadratic programming problem with nonlinear constraints. Then, by using the nonlinear complementarity problem (NCP) function and penalty method this problem is transformed into an unconstrained optimization problem. It is shown that the Hessian matrix of the associated function in the unconstrained optimization problem is positive definite in the optimal point. To solve the unconstrained optimization problem an artificial neural network is used. The proposed neural network has a simple structure and a low complexity of implementation. It is shown here that the proposed artificial neural network is stable in the sense of Lyapunov. Finally, some numerical examples are given to show that the proposed model finds the optimal solution of this problem in the low convergence time. [ABSTRACT FROM AUTHOR]

In this paper, a decentralized fault tolerant control (DFTC) scheme is proposed for a class of large-scale nonlinear systems based on self-tuned local feedback gain against partial loss of actuator effectiveness (PLOAE). Consider a large-scale nonlinear system as a set of interconnected subsystems, a decentralized control method is proposed by employing two radial basis function neural networks (RBFNNs) for the fault-free system. Then, the unknown system is identified using RBFNNs. By establishing a decentralized observer, the derived self-tuned local feedback gain is placed before the proposed decentralized controller to guarantee control performance for the subsystem suffering from PLOAE fault. Finally, simulation examples are provided to demonstrate the effectiveness of the present DFTC scheme. The main contributions of this paper are: i) The unknown large-scale nonlinear system can be identified using locally measured states, so the actuator fault can be handled in its local subsystem. It implies that the performance degradation of the faulty subsystem cannot affect the fault-free subsystems. ii) The estimated effectiveness factor is placed before the proposed decentralized scheme. The fault tolerant control structure is simple since it does not need to be redesigned in the case of PLOAE. [ABSTRACT FROM AUTHOR]

How to extract proper features is very important for synthetic aperture radar (SAR) target configuration recognition. However, most of feature extraction methods are hand-designed and usually can not achieve a satisfactory performance. In this paper, we propose a novel method based on the biologically inspired model to extract features automatically from limited data. Specifically, we learn episodic features (containing the key components and their spatial relations) and semantic features (i.e., semantic descriptions of the key components) which are two important types of features for the human cognition process. Episode features are learned through a deep neural network (DNN) and then semantic geometric features of the key components are defined. Moreover, SAR images are very sensitive to aspect angles. Therefore, we use episode features to estimate aspect angles of testing samples for the final recognition. This paper is a preliminary study and the preliminary experimental results on the moving and stationary target automatic recognition (MSTAR) database demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

In this paper, the exponential stability for a class of neural networks with time-varying delay is concerned. An improved integral inequality is derived which extends the auxiliary function-based integral inequality. A novel Lyapounov-Krasovskii functional (LKF) with some new integral terms is constructed. Based on the improved integral inequality and reciprocally convex combination approach, a less conservative exponential stability criterion for the neural networks with time-varying delay is obtained. The effectiveness of the proposed method in this paper is illustrated via numerical examples. [ABSTRACT FROM AUTHOR]

This paper presents an approach of a social emotional model, which allows to extract the social emotion of a group of intelligent entities. The emotional model PAD allows to represent the emotion of an intelligent entity in 3-D space, allowing the representation of different emotional states. The social emotional model presented in this paper uses individual emotions of each one of the entities, which are represented in the emotional space PAD . Using a social emotional model within intelligent entities allows the creation of more real simulations, in which emotional states can influence decision-making. The result of this social emotional mode is represented by a series of examples, which are intended to represent a number of situations in which the emotions of each individual modify the emotion of the group. Moreover, the paper introduces an example which employs the proposed model in order to learn and predict future actions trying to influence in the social emotion of a group of people. [ABSTRACT FROM AUTHOR]

The attention of this paper is mainly concentrated on the H ∞ state estimator's design problem for a kind of discrete-time memristive neural networks (MNNs) with multiple fading measurements. The phenomenon of multiple fading measurements is represented by a set of individual stochastic variables obeying a predetermined distribution on interval [0,1]. Firstly, the augmented system comprised of MNNs and the dynamics of estimation errors are put forward to implement the performance analysis. Then, under the framework of the difference inclusion theory combined with the Lyapunov function method, several sufficient conditions are established to guarantee the exponential mean-square stability as well as H ∞ performance index. Furthermore, the desired estimator parameter is obtained in view of the solution of a convex optimization problem. In the end, an illustrative numerical example is exploited to check the reliability and usefulness of the design scheme in this paper. [ABSTRACT FROM AUTHOR]

This paper addresses the trajectory-tracking control problem of mobile robot systems with nonholonomic constraints, in the presence of time-varying parametric uncertainties and external disturbances. This necessitates an accurate controller to meet as much control objectives, as possible. Therefore, this paper deals with an artificial intelligence control technique to design a robust controller to meet the control objectives. The design of the intelligent controller is based on the optimal control theory, the adaptive neural network system, and the robust control technique. The trajectory-tracking problem is solved using the optimal control methodology. Since the nonholonomic wheeled mobile robot is strongly nonlinear, the neural network system is applied to approximate the nonlinear function in the optimal control law. The robust controller, for his part, is then applied to adaptively estimate an unknown upper bound of the time-varying parametric uncertainties, external disturbances and approximation error of the neural network system. The stability of the closed-loop robot system is proven using the optimal control theory and Lyapunov stability analysis. The results of the simulation studies on three typical nonholonomic mobile robots are provided to demonstrate the effectiveness of the proposed controller. In addition, a comparative study with a recent robust adaptive controller shows that our proposed intelligent controller gives better results, in the sense that the output trajectory converges to the steady state faster with smaller tracking error. [ABSTRACT FROM AUTHOR]

In this paper, a new design problem of the H ∞ state estimator is developed for a kind of artificial neural networks (ANNs), where both infinite distributed delays and redundant channels are happening. These adopted redundant channels can effectively improve the reliability of networked systems from the viewpoint of engineering. Two series of stochastic variables satisfying Bernoulli distribution, are introduced to govern the infinite distributed delays and schedule the redundant channels. By utilizing both stochastic analysis and Lyapunov functional approach, we obtain a lot of sufficient conditions to ensure the desired H ∞ performance, while the mean-square stability is also satisfied for this investigated state estimation issues of ANNs. The needed estimator gains are designed making use of the matrix inequalities’ solution. In final, a simulation is showed to demonstrate the effectiveness and usefulness of the developed state estimator in this paper. [ABSTRACT FROM AUTHOR]

This paper is concerned with the passivity problem for neural networks with leakage delay and time-varying transmission delay. Using some novel Lyapunov-Krasovskii (L-K) functionals, and combining with the Wirtinger integral inequality, some less conservative delay-dependent passivity conditions are proposed in terms of linear matrix inequalities (LMIs). Different from the existing approaches, the model transformation is abandoned in this paper, and the proposed L-K functionals incorporate the interconnected terms related to both the leakage delay and the time-varying delay. Numerical examples show that the obtained conditions are less conservative. [ABSTRACT FROM AUTHOR]

This paper concerns globally exponential synchronization for a class of impulsive networks with time-delays. The strategy is based on pinning control: single controller is designed to pin the whole network and realize synchronization. Some new conditions are imposed on impulsive sequences, which make the results obtained in this paper less conservative. A numerical example is supplied to illustrate the validity of the main results. [ABSTRACT FROM AUTHOR]

This paper concentrates on the extended dissipativity of memristive neural networks with two additive time-varying delays. After giving a foundation to the memristive model, the paper establishes some fundamental results on quadratically stability and extended dissipativity criteria by means of the Lyapunov functional, integral inequality, as well as the relationship between time-varying delays. The novel extended dissipative inequality contains several weighting matrices, by converting the weighting matrices in a new performance index, the extended dissipativity will be degraded to the H ∞ performance, L 2 − L ∞ performance, passivity and dissipativity, respectively. Finally, one example is given to substantiate the significant improvement of the theoretical approaches. [ABSTRACT FROM AUTHOR]

In this paper, energy consumption of an office building is predicted based on echo state networks (ESNs). Energy consumption of the office building is divided into consumptions from sockets, lights and air-conditioners, which are measured in each room of the office building by three ammeters installed inside, respectively. On the other hand, an office building generally consists of several types of rooms, i.e., office rooms, computer rooms, storage rooms, meeting rooms, etc., the energy consumption of which varies in accordance with different working routines in each type of rooms. In this paper, several novel reservoir topologies of ESNs are developed, the performance of ESNs with different reservoir topologies in predicting the energy consumption of rooms in the office building is compared, and the energy consumption of all the rooms in the office building is predicted with the developed topologies. Moreover, parameter sensitivity of ESNs with different reservoir topologies is analyzed. A case study shows that the developed simplified reservoir topologies are sufficient to achieve outstanding performance of ESNs in the prediction of building energy consumption. [ABSTRACT FROM AUTHOR]

This survey makes an overview of the most recent applications on the neural networks for the computer-aided medical diagnosis (CAMD) over the past decade. CAMD can facilitate the automation of decision making, extraction and visualization of complex characteristics for clinical diagnosis purposes. Over the past decade, neural networks have attained considerable research interest and are widely employed to complex CAMD systems in diverse clinical application domains, such as detecting diseases, classification of diseases, testing the compatibility of new drugs, etc. Overall, this paper reviews the state-of-the-art of neural networks for CAMD. It helps the readers understand the topic of neural networks for CAMD by summarizing the findings addressed in recent academic papers as well as presenting a few open issues of developing the research on this topic. [ABSTRACT FROM AUTHOR]

Very recently, vehicle detection in satellite images has become an emerging research topic with various applications ranging from military to commercial systems. However, it retains as an open problem, mainly due to the complex variations in imaging conditions, object intra-class changes, as well as due to its low-resolution. Coming with the rapid advances in deep learning for feature representation, in this paper we investigate the possibility to exploit deep neural features towards robust vehicle detection. In addition, along with the rapid growth in the data volume, new classification methodology is also demanded to explicitly handle the intra-class variations. In this paper, we propose a vehicle detection framework, which combines Deep Convolutional Neural Network (DNN) based feature learning with Exemplar-SVMs (E-SVMS) based, robust instance classifier to achieve robust vehicle detection in satellite images. In particular, we adopt DNN to learn discriminative image features, which has a high learning capacity. In our practice, the leverage of DNN has achieve significant performance boost by comparing to a serial of handcraft designed features. In addition, we adopt E-SVMs based robust classifier to further improve the classification robustness, which can be considered as an instance-specific metric learning scheme. By conducting extensive experiments with comparisons to a serial of state-of-the-art and alternative works, we further show that the combination of both schemes can benefit from each other to jointly improve the detection accuracy and effectiveness. [ABSTRACT FROM AUTHOR]

This paper discusses the continuous-time consensus problems for directed multi-agent networks under certain coupling or control. Since agents are driven not only by their neighbors' states but also by their derivatives in many realistic situations, dynamical networks described by singular systems are appropriate to study. Consensus problems for such multi-agent networks are considered when the agents communicate in the presence or absence of time delays. The maximum tolerated time-delay is obtained when the multi-agent network asymptotically reaches consensus. The results of this paper indicate that such multi-agent networks can achieve consensus with a demanding convergent speed through agents' interactions. Numerical examples are given to demonstrate the effectiveness of the theoretical results. [ABSTRACT FROM AUTHOR]

In this paper, we tackle the problem of depth estimation from a single image, which is essential for understanding 3D scene structure and can promote the development of visual applications. Markov Random Field (MRF) related depth estimation methods have attracted extensive attentions due to their ability of building structural relationships on outputs. Almost all of them employed engineered low-level absolute and relative features for constructing MRF. However, the engineered features are not capable of producing accurate depths for various scenes. In this paper, we propose Convolutional Conditional Random Field Network (CCRFN) consisting of feature learning and depth tuning components. In feature learning component, we build two Convolutional Neural Network (CNN) architectures to learn absolute and relative features from raw images. In depth tuning component, the learned features are fed into Conditional Random Field (CRF) to generate the depths of all pixels in an image by optimizing a well-defined energy function. CCRFN has two advantages that (1) it does not need hand-crafted features and (2) it models the depths at individual points as well as the relationship between them in the deep model. Experiments on widely used Make3D dataset show that CCRFN outperforms state-of-the-art methods on three evaluation criteria and generates more accurate depth maps with sharper object boundaries. [ABSTRACT FROM AUTHOR]

This paper presents a comprehensive overview of modelling, simulation and implementation of neural networks, taking into account that two aims have emerged in this area: the improvement of our understanding of the behaviour of the nervous system and the need to find inspiration from it to build systems with the advantages provided by nature to perform certain relevant tasks. The development and evolution of different topics related to neural networks is described (simulators, implementations, and real-world applications) showing that the field has acquired maturity and consolidation, proven by its competitiveness in solving real-world problems. The paper also shows how, over time, artificial neural networks have contributed to fundamental concepts at the birth and development of other disciplines such as Computational Neuroscience, Neuro-engineering, Computational Intelligence and Machine Learning. A better understanding of the human brain is considered one of the challenges of this century, and to achieve it, as this paper goes on to describe, several important national and multinational projects and initiatives are marking the way to follow in neural-network research. [ABSTRACT FROM AUTHOR]