Search

This paper addresses the biological plausibility of both backpropagation (BP) and contrastive Hebbian learning (CHL) used in the Boltzmann machines. The main claim of this paper is that CHL is a general learning algorithm that can be used to steer feedforward networks toward desirable outcomes, and steer them away from undesirable outcomes without any need for the specialized feedback circuit of BP or the symmetric connections used by the Boltzmann machines. After adding perturbations during the learning phase to all the neurons in the network, multiple feedforward outcomes are classified into Hebbian and anti-Hebbian sets based on the network predictions. The algorithm is applied to networks when optimizing a loss objective where BP excels and is also applied to networks with stochastic binary outputs where BP cannot be easily applied. The power of the proposed algorithm lies in its simplicity where both learning and gradient estimation through stochastic binary activations are combined into a single local Hebbian rule. We will also show that both Hebbian and anti-Hebbian correlations are evaluated from the readily available signals that are fundamentally different from CHL used in the Boltzmann machines. We will demonstrate that the new learning paradigm where Hebbian/anti-Hebbian correlations are based on correct/incorrect predictions is a powerful concept that separates this paper from other biologically inspired learning algorithms. [ABSTRACT FROM AUTHOR]

Deep learning (DL) is a new machine learning (ML) methodology that has found successful implementations in many application domains. However, its usage in communications systems has not been well explored. This paper investigates the use of the DL in modulation classification, which is a major task in many communications systems. The DL relies on a massive amount of data and, for research and applications, this can be easily available in communications systems. Furthermore, unlike the ML, the DL has the advantage of not requiring manual feature selections, which significantly reduces the task complexity in modulation classification. In this paper, we use two convolutional neural network (CNN)-based DL models, AlexNet and GoogLeNet. Specifically, we develop several methods to represent modulated signals in data formats with gridlike topologies for the CNN. The impacts of representation on classification performance are also analyzed. In addition, comparisons with traditional cumulant and ML-based algorithms are presented. Experimental results demonstrate the significant performance advantage and application feasibility of the DL-based approach for modulation classification. [ABSTRACT FROM AUTHOR]

Deep neural networks bring in impressive accuracy in various applications, but the success often relies on heavy network architectures. Taking well-trained heavy networks as teachers, classical teacher–student learning paradigm aims to learn a student network that is lightweight yet accurate. In this way, a portable student network with significantly fewer parameters can achieve considerable accuracy, which is comparable to that of a teacher network. However, beyond accuracy, the robustness of the learned student network against perturbation is also essential for practical uses. Existing teacher-student learning frameworks mainly focus on accuracy and compression ratios, but ignore the robustness. In this paper, we make the student network produce more confident predictions with the help of the teacher network, and analyze the lower bound of the perturbation that will destroy the confidence of the student network. Two important objectives regarding prediction scores and gradients of examples are developed to maximize this lower bound, to enhance the robustness of the student network without sacrificing the performance. Experiments on benchmark data sets demonstrate the efficiency of the proposed approach to learning robust student networks that have satisfying accuracy and compact sizes. [ABSTRACT FROM AUTHOR]

In recent times, sequence-to-sequence (seq2seq) models have gained a lot of popularity and provide state-of-the-art performance in a wide variety of tasks, such as machine translation, headline generation, text summarization, speech-to-text conversion, and image caption generation. The underlying framework for all these models is usually a deep neural network comprising an encoder and a decoder. Although simple encoder–decoder models produce competitive results, many researchers have proposed additional improvements over these seq2seq models, e.g., using an attention-based model over the input, pointer-generation models, and self-attention models. However, such seq2seq models suffer from two common problems: 1) exposure bias and 2) inconsistency between train/test measurement. Recently, a completely novel point of view has emerged in addressing these two problems in seq2seq models, leveraging methods from reinforcement learning (RL). In this survey, we consider seq2seq problems from the RL point of view and provide a formulation combining the power of RL methods in decision-making with seq2seq models that enable remembering long-term memories. We present some of the most recent frameworks that combine the concepts from RL and deep neural networks. Our work aims to provide insights into some of the problems that inherently arise with current approaches and how we can address them with better RL models. We also provide the source code for implementing most of the RL models discussed in this paper to support the complex task of abstractive text summarization and provide some targeted experiments for these RL models, both in terms of performance and training time. [ABSTRACT FROM AUTHOR]

In recent years, the deep reinforcement learning (DRL) algorithms have been developed rapidly and have achieved excellent performance in many challenging tasks. However, due to the complexity of network structure and a large amount of network parameters, the training of deep network is time-consuming, and consequently, the learning efficiency of DRL is limited. In this paper, aiming to speed up the learning process of DRL agent, we propose a novel approximate policy-based accelerated (APA) algorithm from the viewpoint of the error analysis of approximate policy iteration reinforcement learning algorithms. The proposed APA is proven to be convergent even with a more aggressive learning rate, making the DRL agent have a faster learning speed. Furthermore, to combine the accelerated algorithm with deep Q-network (DQN), Double DQN and deep deterministic policy gradient (DDPG), we proposed three novel DRL algorithms: APA-DQN, APA-Double DQN, and APA-DDPG, which demonstrates the adaptability of the accelerated algorithm with DRL algorithms. We have tested the proposed algorithms on both discrete-action and continuous-action tasks. Their superior performance demonstrates their great potential in the practical applications. [ABSTRACT FROM AUTHOR]

Recent years have witnessed the success of deep learning methods in human activity recognition (HAR). The longstanding shortage of labeled activity data inherently calls for a plethora of semisupervised learning methods, and one of the most challenging and common issues with semisupervised learning is the imbalanced distribution of labeled data over classes. Although the problem has long existed in broad real-world HAR applications, it is rarely explored in the literature. In this paper, we propose a semisupervised deep model for imbalanced activity recognition from multimodal wearable sensory data. We aim to address not only the challenges of multimodal sensor data (e.g., interperson variability and interclass similarity) but also the limited labeled data and class-imbalance issues simultaneously. In particular, we propose a pattern-balanced semisupervised framework to extract and preserve diverse latent patterns of activities. Furthermore, we exploit the independence of multi-modalities of sensory data and attentively identify salient regions that are indicative of human activities from inputs by our recurrent convolutional attention networks. Our experimental results demonstrate that the proposed model achieves a competitive performance compared to a multitude of state-of-the-art methods, both semisupervised and supervised ones, with 10% labeled training data. The results also show the robustness of our method over imbalanced, small training data sets. [ABSTRACT FROM AUTHOR]

Heterogeneous domain adaptation (HDA) aims to solve the learning problems where the source- and the target-domain data are represented by heterogeneous types of features. The existing HDA approaches based on matrix completion or matrix factorization have proven to be effective to capture shareable information between heterogeneous domains. However, there are two limitations in the existing methods. First, a large number of corresponding data instances between the source domain and the target domain are required to bridge the gap between different domains for performing matrix completion. These corresponding data instances may be difficult to collect in real-world applications due to the limited size of data in the target domain. Second, most existing methods can only capture linear correlations between features and data instances while performing matrix completion for HDA. In this paper, we address these two issues by proposing a new matrix-factorization-based HDA method in a semisupervised manner, where only a few labeled data are required in the target domain without requiring any corresponding data instances between domains. Such labeled data are more practical to obtain compared with cross-domain corresponding data instances. Our proposed algorithm is based on matrix factorization in an approximated reproducing kernel Hilbert space (RKHS), where nonlinear correlations between features and data instances can be exploited to learn heterogeneous features for both the source and the target domains. Extensive experiments are conducted on cross-domain text classification and object recognition, and experimental results demonstrate the superiority of our proposed method compared with the state-of-the-art HDA approaches. [ABSTRACT FROM AUTHOR]

Probability density forecast offers the whole distributions of forecasting targets, which brings greater flexibility and practicability than the other probabilistic forecast models such as prediction interval (PI) and quantile forecast. However, existing density forecast models have introduced various constraints on forecasted distributions, which has limited their ability to approximate real distributions and may result in suboptimality. In this paper, a distribution-free density forecast model based on deep learning is proposed, in which the real cumulative density functions (CDFs) of forecasting target are approximated by a large-capacity positive-weighted deep neural network (NN). Benefiting from the universal approximation ability of NNs, the range of forecasted distributions has been proven to contain all the distributions with continuous CDFs, which is superior to existing models’ considering both width and accordance with reality. Three tests from different scenarios were implemented for evaluation, i.e., very-short-term wind power, wind speed, and day-ahead electricity price forecast, in which the proposed density forecast model has shown superior performance over the state of the art. [ABSTRACT FROM AUTHOR]

As nontechnical losses in power systems have recently become a global concern, electricity fraud detection models attracted increasing academic interest. The wide application of smart meters has offered more possibility to detecting fraud from user’s consumption patterns. However, the performances of existing consumption-based electricity fraud detection models are still not satisfactory enough for practice, partly due to their limited ability to handle high-dimensional data. In this paper, a deep-learning-based model is developed for detecting electricity fraud in the advanced metering infrastructure, namely, the multitask feature extracting fraud detector (MFEFD). The deep architecture has brought MFEFD a powerful ability to handle high-dimensional input, through which consumption patterns inside load profiles can be effectively extracted. Another challenge is that the insufficiency of labeled data has restricted the generalization of existing models since they are mostly based on supervised learning and labeled data. MFEFD is trained in a semisupervised manner, in which multitask training was implemented to combine the supervised and unsupervised training, so that both the knowledge from unlabeled and labeled data can be made use of. Real-world-data-based case studies have demonstrated MFEFD’s high detection performance, robustness, privacy preservation, and practicability. [ABSTRACT FROM AUTHOR]

Due to object detection’s close relationship with video analysis and image understanding, it has attracted much research attention in recent years. Traditional object detection methods are built on handcrafted features and shallow trainable architectures. Their performance easily stagnates by constructing complex ensembles that combine multiple low-level image features with high-level context from object detectors and scene classifiers. With the rapid development in deep learning, more powerful tools, which are able to learn semantic, high-level, deeper features, are introduced to address the problems existing in traditional architectures. These models behave differently in network architecture, training strategy, and optimization function. In this paper, we provide a review of deep learning-based object detection frameworks. Our review begins with a brief introduction on the history of deep learning and its representative tool, namely, the convolutional neural network. Then, we focus on typical generic object detection architectures along with some modifications and useful tricks to improve detection performance further. As distinct specific detection tasks exhibit different characteristics, we also briefly survey several specific tasks, including salient object detection, face detection, and pedestrian detection. Experimental analyses are also provided to compare various methods and draw some meaningful conclusions. Finally, several promising directions and tasks are provided to serve as guidelines for future work in both object detection and relevant neural network-based learning systems. [ABSTRACT FROM AUTHOR]

Despite the promising progress made in recent years, person reidentification (re-ID) remains a challenging task due to the complex variations in human appearances from different camera views. This paper proposes to tackle this task by jointly learning feature representation and distance metric in an end-to-end manner. Existing deep metric learning-based re-ID methods usually encounter the following two weaknesses: 1) most works based on pairwise or triplet constraints often suffer from slow convergence and poor local optima, partially because they use very limited samples for each update and 2) hard negative sample mining has been widely applied in existing works. However, hard positive samples, which also contribute to the training of network, have not received enough attention. To alleviate these problems, we develop a novel structural metric learning objective for person re-ID, in which each positive pair is allowed to be compared against all negative pairs in a minibatch and each positive pair is adaptively assigned a hardness-aware weight to modulate its contribution. The introduced positive pair weighting strategy enables the algorithm to focus more on the hard positive samples. Furthermore, we propose to enhance the proposed loss function by adding a global loss term to reduce the variances of positive/negative pair distances, which is able to improve the generalization capability of the network model. By this approach, person images can be nonlinearly mapped into a low-dimensional embedding space where similar samples are kept closer and dissimilar samples are pushed farther apart. We implement the proposed algorithm using the inception architecture and evaluate it on three large-scale re-ID data sets. Experiment results demonstrate that our approach is able to outperform most state of the arts while using much lower dimensional deep features. [ABSTRACT FROM AUTHOR]

This paper proposes a new and efficient technique to regularize the neural network in the context of deep learning using correlations among features. Previous studies have shown that oversized deep neural network models tend to produce a lot of redundant features that are either the shifted version of one another or are very similar and show little or no variations, thus resulting in redundant filtering. We propose a way to address this problem and show that such redundancy can be avoided using regularization and adaptive feature dropout mechanism. We show that regularizing both negative and positive correlated features according to their differentiation and based on their relative cosine distances yields network extracting dissimilar features with less overfitting and better generalization. This concept is illustrated with deep multilayer perceptron, convolutional neural network, sparse autoencoder, gated recurrent unit, and long short-term memory on MNIST digits recognition, CIFAR-10, ImageNet, and Stanford Natural Language Inference data sets. [ABSTRACT FROM AUTHOR]

Explicitly or implicitly, most dimensionality reduction methods need to determine which samples are neighbors and the similarities between the neighbors in the original high-dimensional space. The projection matrix is then learnt on the assumption that the neighborhood information, e.g., the similarities, are known and fixed prior to learning. However, it is difficult to precisely measure the intrinsic similarities of samples in high-dimensional space because of the curse of dimensionality. Consequently, the neighbors selected according to such similarities and the projection matrix obtained according to such similarities and the corresponding neighbors might not be optimal in the sense of classification and generalization. To overcome this drawback, in this paper, we propose to let the similarities and neighbors be variables and model these in a low-dimensional space. Both the optimal similarity and projection matrix are obtained by minimizing a unified objective function. Nonnegative and sum-to-one constraints on the similarity are adopted. Instead of empirically setting the regularization parameter, we treat it as a variable to be optimized. It is interesting that the optimal regularization parameter is adaptive to the neighbors in a low-dimensional space and has an intuitive meaning. Experimental results on the YALE B, COIL-100, and MNIST data sets demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Batch normalization (BN) has recently become a standard component for accelerating and improving the training of deep neural networks (DNNs). However, BN brings in additional calculations, consumes more memory, and significantly slows down the training iteration. Furthermore, the nonlinear square and sqrt operations in the normalization process impede low bit-width quantization techniques, which draw much attention to the deep learning hardware community. In this paper, we propose an $L1$ -norm BN (L1BN) with only linear operations in both forward and backward propagations during training. L1BN is approximately equivalent to the conventional $L2$ -norm BN (L2BN) by multiplying a scaling factor that equals $({\pi }/{2})^{1/2}$. Experiments on various convolutional neural networks and generative adversarial networks reveal that L1BN can maintain the same performance and convergence rate as L2BN but with higher computational efficiency. In real application-specified integrated circuit synthesis with reduced resources, L1BN achieves 25% speedup and 37% energy saving compared to the original L2BN. Our hardware-friendly normalization method not only surpasses L2BN in speed but also simplifies the design of deep learning accelerators. Last but not least, L1BN promises a fully quantized training of DNNs, which empowers future artificial intelligence applications on mobile devices with transfer and continual learning capability. [ABSTRACT FROM AUTHOR]

Generative adversarial networks (GANs) learn a deep generative model that is able to synthesize novel, high-dimensional data samples. New data samples are synthesized by passing latent samples, drawn from a chosen prior distribution, through the generative model. Once trained, the latent space exhibits interesting properties that may be useful for downstream tasks such as classification or retrieval. Unfortunately, GANs do not offer an “inverse model,” a mapping from data space back to latent space, making it difficult to infer a latent representation for a given data sample. In this paper, we introduce a technique, inversion, to project data samples, specifically images, to the latent space using a pretrained GAN. Using our proposed inversion technique, we are able to identify which attributes of a data set a trained GAN is able to model and quantify GAN performance, based on a reconstruction loss. We demonstrate how our proposed inversion technique may be used to quantitatively compare the performance of various GAN models trained on three image data sets. We provide codes for all of our experiments in the website (https://github.com/ToniCreswell/InvertingGAN). [ABSTRACT FROM AUTHOR]

An efficient deep learning requires a memory-efficient construction of a neural network. This paper introduces a layerwise tensorized formulation of a multilayer neural network, called LTNN, such that the weight matrix can be significantly compressed during training. By reshaping the multilayer neural network weight matrix into a high-dimensional tensor with a low-rank approximation, significant network compression can be achieved with maintained accuracy. An according layerwise training is developed by a modified alternating least-squares method with backward propagation for fine-tuning only. LTNN can provide the state-of-the-art results on various benchmarks with significant compression. For MNIST benchmark, LTNN shows $64 \times $ compression rate without accuracy drop. For Imagenet12 benchmark, our proposed LTNN achieves $35.84 \times $ compression of the neural network with around 2% accuracy drop. We have also shown $1.615 \times $ faster on inference speed than the existing works due to the smaller tensor core ranks. [ABSTRACT FROM AUTHOR]

Image recognition based on convolutional neural networks (CNNs) has recently been shown to deliver the state-of-the-art performance in various areas of computer vision and image processing. Nevertheless, applying a deep CNN to no-reference image quality assessment (NR-IQA) remains a challenging task due to critical obstacles, i.e., the lack of a training database. In this paper, we propose a CNN-based NR-IQA framework that can effectively solve this problem. The proposed method—deep image quality assessor (DIQA)—separates the training of NR-IQA into two stages: 1) an objective distortion part and 2) a human visual system-related part. In the first stage, the CNN learns to predict the objective error map, and then the model learns to predict subjective score in the second stage. To complement the inaccuracy of the objective error map prediction on the homogeneous region, we also propose a reliability map. Two simple handcrafted features were additionally employed to further enhance the accuracy. In addition, we propose a way to visualize perceptual error maps to analyze what was learned by the deep CNN model. In the experiments, the DIQA yielded the state-of-the-art accuracy on the various databases. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel approach for efficient training of deep neural networks in a bottom-up fashion using a layered structure. Our algorithm, which we refer to as deep cascade learning, is motivated by the cascade correlation approach of Fahlman and Lebiere, who introduced it in the context of perceptrons. We demonstrate our algorithm on networks of convolutional layers, though its applicability is more general. Such training of deep networks in a cascade directly circumvents the well-known vanishing gradient problem by ensuring that the output is always adjacent to the layer being trained. We present empirical evaluations comparing our deep cascade training with standard end–end training using back propagation of two convolutional neural network architectures on benchmark image classification tasks (CIFAR-10 and CIFAR-100). We then investigate the features learned by the approach and find that better, domain-specific, representations are learned in early layers when compared to what is learned in end–end training. This is partially attributable to the vanishing gradient problem that inhibits early layer filters to change significantly from their initial settings. While both networks perform similarly overall, recognition accuracy increases progressively with each added layer, with discriminative features learned in every stage of the network, whereas in end–end training, no such systematic feature representation was observed. We also show that such cascade training has significant computational and memory advantages over end–end training, and can be used as a pretraining algorithm to obtain a better performance. [ABSTRACT FROM AUTHOR]

In this paper, we propose a new end-to-end deep neural network model for time-series classification (TSC) with emphasis on both the accuracy and the interpretation. The proposed model contains a convolutional network component to extract high-level features and a recurrent network component to enhance the modeling of the temporal characteristics of TS data. In addition, a feedforward fully connected network with the sparse group lasso (SGL) regularization is used to generate the final classification. The proposed architecture not only achieves satisfying classification accuracy, but also obtains good interpretability through the SGL regularization. All these networks are connected and jointly trained in an end-to-end framework, and it can be generally applied to TSC tasks across different domains without the efforts of feature engineering. Our experiments in various TS data sets show that the proposed model outperforms the traditional convolutional neural network model for the classification accuracy, and also demonstrate how the SGL contributes to a better model interpretation. [ABSTRACT FROM AUTHOR]

Unsupervised feature extractors are known to perform an efficient and discriminative representation of data. Insight into the mappings they perform and human ability to understand them, however, remain very limited. This is especially prominent when multilayer deep learning architectures are used. This paper demonstrates how to remove these bottlenecks within the architecture of non-negativity constrained autoencoder. It is shown that using both L1 and L2 regularizations that induce non-negativity of weights, most of the weights in the network become constrained to be non-negative, thereby resulting into a more understandable structure with minute deterioration in classification accuracy. Also, this proposed approach extracts features that are more sparse and produces additional output layer sparsification. The method is analyzed for accuracy and feature interpretation on the MNIST data, the NORB normalized uniform object data, and the Reuters text categorization data set. [ABSTRACT FROM AUTHOR]

Learning deep representations have been applied in action recognition widely. However, there have been a few investigations on how to utilize the structural manifold information among different action videos to enhance the recognition accuracy and efficiency. In this paper, we propose to incorporate the manifold of training samples into deep learning, which is defined as deep manifold learning (DML). The proposed DML framework can be adapted to most existing deep networks to learn more discriminative features for action recognition. When applied to a convolutional neural network, DML embeds the previous convolutional layer’s manifold into the next convolutional layer; thus, the discriminative capacity of the next layer can be promoted. We also apply the DML on a restricted Boltzmann machine, which can alleviate the overfitting problem. Experimental results on four standard action databases (i.e., UCF101, HMDB51, KTH, and UCF sports) show that the proposed method outperforms the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Extreme learning machine (ELM) is an emerging learning algorithm for the generalized single hidden layer feedforward neural networks, of which the hidden node parameters are randomly generated and the output weights are analytically computed. However, due to its shallow architecture, feature learning using ELM may not be effective for natural signals (e.g., images/videos), even with a large number of hidden nodes. To address this issue, in this paper, a new ELM-based hierarchical learning framework is proposed for multilayer perceptron. The proposed architecture is divided into two main components: 1) self-taught feature extraction followed by supervised feature classification and 2) they are bridged by random initialized hidden weights. The novelties of this paper are as follows: 1) unsupervised multilayer encoding is conducted for feature extraction, and an ELM-based sparse autoencoder is developed via \ell 1 constraint. By doing so, it achieves more compact and meaningful feature representations than the original ELM; 2) by exploiting the advantages of ELM random feature mapping, the hierarchically encoded outputs are randomly projected before final decision making, which leads to a better generalization with faster learning speed; and 3) unlike the greedy layerwise training of deep learning (DL), the hidden layers of the proposed framework are trained in a forward manner. Once the previous layer is established, the weights of the current layer are fixed without fine-tuning. Therefore, it has much better learning efficiency than the DL. Extensive experiments on various widely used classification data sets show that the proposed algorithm achieves better and faster convergence than the existing state-of-the-art hierarchical learning methods. Furthermore, multiple applications in computer vision further confirm the generality and capability of the proposed learning scheme. [ABSTRACT FROM AUTHOR]

This paper investigates how to blindly evaluate the visual quality of an image by learning rules from linguistic descriptions. Extensive psychological evidence shows that humans prefer to conduct evaluations qualitatively rather than numerically. The qualitative evaluations are then converted into the numerical scores to fairly benchmark objective image quality assessment (IQA) metrics. Recently, lots of learning-based IQA models are proposed by analyzing the mapping from the images to numerical ratings. However, the learnt mapping can hardly be accurate enough because some information has been lost in such an irreversible conversion from the linguistic descriptions to numerical scores. In this paper, we propose a blind IQA model, which learns qualitative evaluations directly and outputs numerical scores for general utilization and fair comparison. Images are represented by natural scene statistics features. A discriminative deep model is trained to classify the features into five grades, corresponding to five explicit mental concepts, i.e., excellent, good, fair, poor, and bad. A newly designed quality pooling is then applied to convert the qualitative labels into scores. The classification framework is not only much more natural than the regression-based models, but also robust to the small sample size problem. Thorough experiments are conducted on popular databases to verify the model’s effectiveness, efficiency, and robustness. [ABSTRACT FROM AUTHOR]

Deep neural networks (NNs) are the state-of-the-art models for understanding the content of images and videos. However, implementing deep NNs in embedded systems is a challenging task, e.g., a typical deep belief network could exhaust gigabytes of memory and result in bandwidth and computational bottlenecks. To address this challenge, this paper presents an algorithm and hardware codesign for efficient deep neural computation. A hardware-oriented deep learning algorithm, named the deep adaptive network, is proposed to explore the sparsity of neural connections. By adaptively removing the majority of neural connections and robustly representing the reserved connections using binary integers, the proposed algorithm could save up to 99.9% memory utility and computational resources without undermining classification accuracy. An efficient sparse-mapping-memory-based hardware architecture is proposed to fully take advantage of the algorithmic optimization. Different from traditional Von Neumann architecture, the deep-adaptive network on chip (DANoC) brings communication and computation in close proximity to avoid power-hungry parameter transfers between on-board memory and on-chip computational units. Experiments over different image classification benchmarks show that the DANoC system achieves competitively high accuracy and efficiency comparing with the state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Deep reinforcement learning (RL) comprehensively uses the psychological mechanisms of “trial and error” and “reward and punishment” in RL as well as powerful feature expression and nonlinear mapping in deep learning. Currently, it plays an essential role in the fields of artificial intelligence and machine learning. Since an RL agent needs to constantly interact with its surroundings, the deep Q network (DQN) is inevitably faced with the need to learn numerous network parameters, which results in low learning efficiency. In this paper, a multisource transfer double DQN (MTDDQN) based on actor learning is proposed. The transfer learning technique is integrated with deep RL to make the RL agent collect, summarize, and transfer action knowledge, including policy mimic and feature regression, to the training of related tasks. There exists action overestimation in DQN, i.e., the lower probability limit of action corresponding to the maximum Q value is nonzero. Therefore, the transfer network is trained by using double DQN to eliminate the error accumulation caused by action overestimation. In addition, to avoid negative transfer, i.e., to ensure strong correlations between source and target tasks, a multisource transfer learning mechanism is applied. The Atari2600 game is tested on the arcade learning environment platform to evaluate the feasibility and performance of MTDDQN by comparing it with some mainstream approaches, such as DQN and double DQN. Experiments prove that MTDDQN achieves not only human-like actor learning transfer capability, but also the desired learning efficiency and testing accuracy on target task. [ABSTRACT FROM AUTHOR]

In this paper, a new training paradigm is proposed for deep reinforcement learning using self-paced prioritized curriculum learning with coverage penalty. The proposed deep curriculum reinforcement learning (DCRL) takes the most advantage of experience replay by adaptively selecting appropriate transitions from replay memory based on the complexity of each transition. The criteria of complexity in DCRL consist of self-paced priority as well as coverage penalty. The self-paced priority reflects the relationship between the temporal-difference error and the difficulty of the current curriculum for sample efficiency. The coverage penalty is taken into account for sample diversity. With comparison to deep Q network (DQN) and prioritized experience replay (PER) methods, the DCRL algorithm is evaluated on Atari 2600 games, and the experimental results show that DCRL outperforms DQN and PER on most of these games. More results further show that the proposed curriculum training paradigm of DCRL is also applicable and effective for other memory-based deep reinforcement learning approaches, such as double DQN and dueling network. All the experimental results demonstrate that DCRL can achieve improved training efficiency and robustness for deep reinforcement learning. [ABSTRACT FROM AUTHOR]

This paper proposes a novel method, called robust latent subspace learning (RLSL), for image classification. We formulate an RLSL problem as a joint optimization problem over both the latent SL and classification model parameter predication, which simultaneously minimizes: 1) the regression loss between the learned data representation and objective outputs and 2) the reconstruction error between the learned data representation and original inputs. The latent subspace can be used as a bridge that is expected to seamlessly connect the origin visual features and their class labels and hence improve the overall prediction performance. RLSL combines feature learning with classification so that the learned data representation in the latent subspace is more discriminative for classification. To learn a robust latent subspace, we use a sparse item to compensate error, which helps suppress the interference of noise via weakening its response during regression. An efficient optimization algorithm is designed to solve the proposed optimization problem. To validate the effectiveness of the proposed RLSL method, we conduct experiments on diverse databases and encouraging recognition results are achieved compared with many state-of-the-arts methods. [ABSTRACT FROM AUTHOR]

The growing interests in multiway data analysis and deep learning have drawn tensor factorization (TF) and neural network (NN) as the crucial topics. Conventionally, the NN model is estimated from a set of one-way observations. Such a vectorized NN is not generalized for learning the representation from multiway observations. The classification performance using vectorized NN is constrained, because the temporal or spatial information in neighboring ways is disregarded. More parameters are required to learn the complicated data structure. This paper presents a new tensor-factorized NN (TFNN), which tightly integrates TF and NN for multiway feature extraction and classification under a unified discriminative objective. This TFNN is seen as a generalized NN, where the affine transformation in an NN is replaced by the multilinear and multiway factorization for tensor-based NN. The multiway information is preserved through layerwise factorization. Tucker decomposition and nonlinear activation are performed in each hidden layer. The tensor-factorized error backpropagation is developed to train TFNN with the limited parameter size and computation time. This TFNN can be further extended to realize the convolutional TFNN (CTFNN) by looking at small subtensors through the factorized convolution. Experiments on real-world classification tasks demonstrate that TFNN and CTFNN attain substantial improvement when compared with an NN and a convolutional NN, respectively. [ABSTRACT FROM AUTHOR]

Although there have been many decades of research and commercial presence on high performance general purpose processors, there are still many applications that require fully customized hardware architectures for further computational acceleration. Recently, deep learning has been successfully used to learn in a wide variety of applications, but their heavy computation demand has considerably limited their practical applications. This paper proposes a fully pipelined acceleration architecture to alleviate high computational demand of an artificial neural network (ANN) which is restricted Boltzmann machine (RBM) ANNs. The implemented RBM ANN accelerator (integrating $1024\times 1024$ network size, using 128 input cases per batch, and running at a 303-MHz clock frequency) integrated in a state-of-the art field-programmable gate array (FPGA) (Xilinx Virtex 7 XC7V-2000T) provides a computational performance of 301-billion connection-updates-per-second and about 193 times higher performance than a software solution running on general purpose processors. Most importantly, the architecture enables over 4 times (12 times in batch learning) higher performance compared with a previous work when both are implemented in an FPGA device (XC2VP70). [ABSTRACT FROM AUTHOR]

Developments in deep learning have seen the use of layerwise unsupervised learning combined with supervised learning for fine-tuning. With this layerwise approach, a deep network can be seen as a more modular system that lends itself well to learning representations. In this paper, we investigate whether such modularity can be useful to the insertion of background knowledge into deep networks, whether it can improve learning performance when it is available, and to the extraction of knowledge from trained deep networks, and whether it can offer a better understanding of the representations learned by such networks. To this end, we use a simple symbolic language—a set of logical rules that we call confidence rules—and show that it is suitable for the representation of quantitative reasoning in deep networks. We show by knowledge extraction that confidence rules can offer a low-cost representation for layerwise networks (or restricted Boltzmann machines). We also show that layerwise extraction can produce an improvement in the accuracy of deep belief networks. Furthermore, the proposed symbolic characterization of deep networks provides a novel method for the insertion of prior knowledge and training of deep networks. With the use of this method, a deep neural–symbolic system is proposed and evaluated, with the experimental results indicating that modularity through the use of confidence rules and knowledge insertion can be beneficial to network performance. [ABSTRACT FROM AUTHOR]

Broad Learning System (BLS) that aims to offer an alternative way of learning in deep structure is proposed in this paper. Deep structure and learning suffer from a time-consuming training process because of a large number of connecting parameters in filters and layers. Moreover, it encounters a complete retraining process if the structure is not sufficient to model the system. The BLS is established in the form of a flat network, where the original inputs are transferred and placed as “mapped features” in feature nodes and the structure is expanded in wide sense in the “enhancement nodes.” The incremental learning algorithms are developed for fast remodeling in broad expansion without a retraining process if the network deems to be expanded. Two incremental learning algorithms are given for both the increment of the feature nodes (or filters in deep structure) and the increment of the enhancement nodes. The designed model and algorithms are very versatile for selecting a model rapidly. In addition, another incremental learning is developed for a system that has been modeled encounters a new incoming input. Specifically, the system can be remodeled in an incremental way without the entire retraining from the beginning. Satisfactory result for model reduction using singular value decomposition is conducted to simplify the final structure. Compared with existing deep neural networks, experimental results on the Modified National Institute of Standards and Technology database and NYU NORB object recognition dataset benchmark data demonstrate the effectiveness of the proposed BLS. [ABSTRACT FROM AUTHOR]

A major progress in deep multilayer neural networks (DNNs) is the invention of various unsupervised pretraining methods to initialize network parameters which lead to good prediction accuracy. This paper presents the sparseness analysis on the hidden unit in the pretraining process. In particular, we use the L1 -norm to measure sparseness and provide some sufficient conditions for that pretraining leads to sparseness with respect to the popular pretraining models—such as denoising autoencoders (DAEs) and restricted Boltzmann machines (RBMs). Our experimental results demonstrate that when the sufficient conditions are satisfied, the pretraining models lead to sparseness. Our experiments also reveal that when using the sigmoid activation functions, pretraining plays an important sparseness role in DNNs with sigmoid (Dsigm), and when using the rectifier linear unit (ReLU) activation functions, pretraining becomes less effective for DNNs with ReLU (Drelu). Luckily, Drelu can reach a higher recognition accuracy than DNNs with pretraining (DAEs and RBMs), as it can capture the main benefit (such as sparseness-encouraging) of pretraining in Dsigm. However, ReLU is not adapted to the different firing rates in biological neurons, because the firing rate actually changes along with the varying membrane resistances. To address this problem, we further propose a family of rectifier piecewise linear units (RePLUs) to fit the different firing rates. The experimental results show that the performance of RePLU is better than ReLU, and is comparable with those with some pretraining techniques, such as RBMs and DAEs. [ABSTRACT FROM PUBLISHER]

Can we train the computer to beat experienced traders for financial assert trading? In this paper, we try to address this challenge by introducing a recurrent deep neural network (NN) for real-time financial signal representation and trading. Our model is inspired by two biological-related learning concepts of deep learning (DL) and reinforcement learning (RL). In the framework, the DL part automatically senses the dynamic market condition for informative feature learning. Then, the RL module interacts with deep representations and makes trading decisions to accumulate the ultimate rewards in an unknown environment. The learning system is implemented in a complex NN that exhibits both the deep and recurrent structures. Hence, we propose a task-aware backpropagation through time method to cope with the gradient vanishing issue in deep training. The robustness of the neural system is verified on both the stock and the commodity future markets under broad testing conditions. [ABSTRACT FROM PUBLISHER]

As an interesting and emerging topic, cosaliency detection aims at simultaneously extracting common salient objects in multiple related images. It differs from the conventional saliency detection paradigm in which saliency detection for each image is determined one by one independently without taking advantage of the homogeneity in the data pool of multiple related images. In this paper, we propose a novel cosaliency detection approach using deep learning models. Two new concepts, called intrasaliency prior transfer and deep intersaliency mining, are introduced and explored in the proposed work. For the intrasaliency prior transfer, we build a stacked denoising autoencoder (SDAE) to learn the saliency prior knowledge from auxiliary annotated data sets and then transfer the learned knowledge to estimate the intrasaliency for each image in cosaliency data sets. For the deep intersaliency mining, we formulate it by using the deep reconstruction residual obtained in the highest hidden layer of a self-trained SDAE. The obtained deep intersaliency can extract more intrinsic and general hidden patterns to discover the homogeneity of cosalient objects in terms of some higher level concepts. Finally, the cosaliency maps are generated by weighted integration of the proposed intrasaliency prior, deep intersaliency, and traditional shallow intersaliency. Comprehensive experiments over diverse publicly available benchmark data sets demonstrate consistent performance gains of the proposed method over the state-of-the-art cosaliency detection methods. [ABSTRACT FROM PUBLISHER]

This paper presents a novel change detection approach for synthetic aperture radar images based on deep learning. The approach accomplishes the detection of the changed and unchanged areas by designing a deep neural network. The main guideline is to produce a change detection map directly from two images with the trained deep neural network. The method can omit the process of generating a difference image (DI) that shows difference degrees between multitemporal synthetic aperture radar images. Thus, it can avoid the effect of the DI on the change detection results. The learning algorithm for deep architectures includes unsupervised feature learning and supervised fine-tuning to complete classification. The unsupervised feature learning aims at learning the representation of the relationships between the two images. In addition, the supervised fine-tuning aims at learning the concepts of the changed and unchanged pixels. Experiments on real data sets and theoretical analysis indicate the advantages, feasibility, and potential of the proposed method. Moreover, based on the results achieved by various traditional algorithms, respectively, deep learning can further improve the detection performance. [ABSTRACT FROM PUBLISHER]