Search

Concerning the problems that the traditional Convolutional Neural Network (CNN) ignores contextual semantic information, and the traditional Recurrent Neural Network (RNN) has information memory loss and vanishing gradient, this paper proposes a Bi‐directional Encoder Representations from Transformers (BERT)‐based dual‐channel parallel hybrid neural network model for text sentiment analysis. The BERT model is used to convert text into word vectors; the dual‐channel parallel hybrid neural network model constructed by CNN and Bi‐directional Long Short‐Term Memory (BiLSTM) extracts local and global semantic features of the text, which can obtain more comprehensive sentiment features; the attention mechanism enables some words to get more attention that highlights important words and improves the model's sentiment classification ability. Finally, the dual‐channel output features are fused for sentiment classification. The experimental results on the hotel review datasets show that the Accuracy of the proposed model in sentiment classification reaches 92.35% and the F1 score reaches 91.59%. [ABSTRACT FROM AUTHOR]

Many real-world classification problems are becoming multi-label in nature, i.e., multiple class labels are assigned to an instance simultaneously. Multi-label classification is a challenging problem due to the involvement of three forms of interactions, i.e., feature-to-feature, feature-to-label, and label-to-label interactions. What further complicates the problem is that not all features are useful, and some can deteriorate the classification performance. Sparsity-based methods have been widely used to address multi-label feature selection due to their efficiency and effectiveness. However, most (if not all) existing methods do not consider the three forms of interactions simultaneously, which could hinder their ability to achieve good performance. Moreover, most existing methods are gradient-based, which are prone to getting stuck at local optima. This paper proposes a new sparsity-based feature selection approach that can simultaneously consider all three forms of interactions. Furthermore, this paper develops a novel sparse learning method based on particle swarm optimisation that can avoid local optima. The proposed method is compared against the state-of-the-art multi-label feature selection methods in terms of multi-label classification performance. The results show that our method performed significantly better in selecting high-quality feature subsets with respect to various feature subset sizes. [ABSTRACT FROM AUTHOR]

Nowadays the robot manipulation skills are usually learned by human demonstration via trajectory-level learning, which somewhat lacks robustness and generalization. In this paper, we propose a novel contact state level learning method for robot manipulation skill acquisition via human demonstration. The robot-environment contact states are described via environment dynamics modelling and geometric constraints modelling for flexible contact and rigid contact cases, respectively. During human demonstration process, the robot-environment interaction force, the robot position, and velocity data are collected. After that, the environment dynamics and geometric constraints modelling methods are presented to determine the contact state changes during the robot manipulation process. Then the robot manipulator learns the contact state information rather than specific manipulation trajectory. On this basis, the manipulation control law using active exploration method is presented to control the robot during the button pressing process and peg-hole-insertion process, respectively. Finally, the performance of the presented methodology has been verified via experimental studies. Note to Practitioners—Intelligent robots will become the right assistants of human beings in the future, especially in various areas of manipulation occasions. The important premise of realizing this vision is that the robots should have certain ability of manipulation skill learning. A lot of research has been carried out in this field, many of which are focusing on trajectory level manipulation skill learning and reproduction. Other than the trajectory level learning, human beings can learn many other higher levels of manipulation skills, such as the contact state level and semantic level learning, which makes the learning results more robust and general. In this paper, the contact state estimation and learning method via environment dynamics and geometric constraints modelling is presented to learn the robot manipulation skill based on the contact state transition conditions. In this way, the robot needs less data in the skill learning process, and the trajectory level learning is avoided. After learning the contact state level manipulation skill, the lower trajectory level command is autonomously generated. Experiments on button pressing and peg-hole-insertion tasks by KUKA iiwa robot have obtained very good results. Other than the button pressing and peg-hole-insertion tasks, the presented methodology can be applied to many other manipulation tasks, as long as there are contact state changes in the manipulation process. The work of this paper lays a foundation for the robot learning of higher-level manipulation skills. [ABSTRACT FROM AUTHOR]

The papers in this special section focus on designing technologies to support professional and workplace learning in situated practices. In an era of global, organizational, and technological change, all of which are transforming the world of work, professional and workplace learning are critical for employability and organizational competitiveness. A range of fundamental transformations is changing how people work. Digital technologies are replacing human labor and, at the same time, are accelerating the expansion of job roles and work practices. Work is becoming increasingly specialized, which means that professionals in collaborative and networked ways across discipline and organization boundaries. In parallel, labor is increasingly decentralized, making decisionmaking more distributed and raising the need for remote communication and collaboration. Subsequently, work is becoming more independent from time and place, as people connect, collaborate, and work via digital technologies. These changes come with a need for substantial and continuous workplace learning, and with the need for changes in how workplace learning happens. Of course, digital technologies are already used to provide learning and training in workplaces. However, most of these learning technologies have been developed for formal education (e.g., K- 12 and higher education) rather than in workplace contexts. There is a need to understand and evidence workplace learning needs and to further develop technologies that can support and scale workplace learning. [ABSTRACT FROM AUTHOR]

Human beings tend to learn unknown knowledge in a gradual process, from the basic to the complex. Based on this point, we propose a progressive learning method for producing service strategies according to requests, with a hierarchical priori knowledge and reinforcement learning. Service strategy aims to guide how to perform home services and takes into consideration the relationship between actions and objects in home environment. In this paper, strategy generation is regarded as a text generation problem in question answering (QA). Firstly, a hierarchical priori knowledge with service-object correlation at the bottom and action-object correlation at the top is constructed to assist the understanding on the relationship of objects and actions in service strategies. Service-object correlation guides how to select proper objects with the correct order, while action-object correlation associates actions in strategies according to selected objects. Based on the hierarchical priori knowledge, a progressive learning method is proposed to make the model produce effective strategies with a sequential cognition, from service-object correlation (objects) to action-object correlation (actions). After that, reinforcement learning is employed to enhance the progressive guidance, by designing rewards in terms of the hierarchical priori knowledge. Finally, the proposed method is tested with both comparative experiments and ablation studies, and the experimental results demonstrate the superiority in producing comprehensive and logical strategies, indicating that the progressive learning method in our paper can further improve the QA performance. [ABSTRACT FROM AUTHOR]

The use of Artificial Intelligence and Big Data in health care opens up new opportunities for the measurement of the human. Their application aims not only at gathering more and better data points but also at doing it less invasive. With this change in health care towards its extension to almost all areas of life and its increasing invisibility and opacity, new questions of transparency arise. While the complex human-machine interactions involved in deploying and using AI tend to become non-transparent, the use of these technologies makes the patient seemingly transparent. Papers on the ethical implementation of AI plead for transparency but neglect the factor of the “transparent patient” as intertwined with AI. Transparency in this regard appears to be Janus-faced: The precondition for receiving help - e.g., treatment advice regarding the own health - is to become transparent for the digitized health care system. That is, for instance, to donate data and become visible to the AI and its operators. The paper reflects on this entanglement of transparent patients and (non-) transparent technology. It argues that transparency regarding both AI and humans is not an ethical principle per se but an infraethical concept. Further, it is no sufficient basis for avoiding harm and human dignity violations. Rather, transparency must be enriched by intelligibility following Judith Butler’s use of the term. Intelligibility is understood as an epistemological presupposition for recognition and the ensuing humane treatment. Finally, the paper highlights ways to testify intelligibility in dealing with AI in health care ex ante, ex post, and continuously. [ABSTRACT FROM AUTHOR]

Due to the lack of distinctiveness to the background, robust monocular 6DOF pose tracking of less-distinct objects remains an important open problem. In this paper, we firstly analyze the object distinctiveness in the tracking process. Then, we propose a novel contour-part model based robust monocular pose tracking method for less-distinct objects. This paper uses the traditional contour feature for pose tracking in a novel strategy called contour part model. First, the contour part model is built by segmenting the projected contour rendered from the 3D model into contour segments of a certain length adaptively according to the Shi-Tomasi cornerness scores. Then, the correspondence is detected in the input image for each contour part by gradient orientation based template matching. Finally, pose tracking is achieved by solving the PnP problem. Experiment results on semi-synthetic and real images show that the proposed method performs better than the existing methods when pose tracking less-distinct objects and shows great robustness toward interference. Additionally, we combine the edge feature and the regional feature in a simple strategy. The overall performance of the fused method is boosted according to the experimental results on the RBOT dataset. The fact shows that the fusion of the edge and regional features has great potential for improving tracking performance. [ABSTRACT FROM AUTHOR]

Brief historical background on the establishment and activities of the Institute of Control Sciences (IPU RAS). The paper presents key findings of the Institute (obtained mainly in the 20th century) in the field of pattern recognition and of related analysis of complex data. It focuses on four areas of research including (a) the method of potential functions, (b) the theory of learning and self-learning systems, (c) the generalized portrait method and recovery of dependences based on empirical data, and (d) automatic classification methods and expert classification analysis. Relations between these areas are studied. The pioneers in the field are named (M.A. Aizerman, E.M. Braverman, L.I. Rozonoer, Ya.Z. Tsypkin, V.N. Vapnik, A.Ya. Chervonenkis, I.B. Muchnik, and A.A. Dorofeyuk among others) and brief biographical notes on the life and scientific work of these scientists are presented. The follow-ups of the results thus obtained are shown. The bibliography of publications by the Institute's researchers in leading journals of Russia on pattern recognition problems and related complex data analysis tasks is provided. [ABSTRACT FROM AUTHOR]

This paper proposes a novel ensemble learning-based adaptive stochastic model predictive control (SMPC) algorithm for constrained linear systems with unknown nonlinear terms and random disturbances. The ensemble network combining a feedforward neural network and a Bayesian network is used to offline learn the nonlinear dynamics and disturbance distribution parameters. Then, the mixed-tube scheme is designed to cope with input constraints and state chance constraints while decreasing computational demands and conservativeness. The reliability of the stochastic tube is guaranteed using the Hoeffding inequality-based verification mechanism, which results in a chance constraint with double probabilities. The feasibility and exponential stability of the SMPC are rigorously proven. A numerical example verifies the merits of the proposed algorithm in terms of the control performance and the feasible domain. [ABSTRACT FROM AUTHOR]

Deep Learning (DL) is becoming more popular in the healthcare sectors due to the exponential growth of data availability and its excellent performance in diagnosing various diseases. This paper has aimed to design a new possible brain tumor diagnostic model to improve accuracy and reliability of radiology. In this paper, an advanced deep learning algorithm is used to detect and classify brain tumors in magnetic resonance (MR) images. Diagnosing brain tumors in radiology is a significant issue, yet it is a difficult and time-consuming procedure that radiologists must pass through. The reliability of their assessment relies completely on their knowledge and personal judgements which are in most cases inaccurate. As a possible remedy to the growing concern in diagnosing brain tumors accurately, in this work a deep learning method is applied to classify the brain tumor MR images with very high performance accuracy. The research leveraged a transfer learning model known as AlexNet's convolutional neural network (CNN) to perform this operation. Our method helps to improve robustness, efficiencies and accuracy in the healthcare sector with the ability to automate the entire diagnostic process with the overall accuracy of 99.62%. Additionally, our model has the ability to detect and classify tumors at their different stages and magnitudes. [ABSTRACT FROM AUTHOR]

Uncertain cut-in maneuver of vehicles from adjacent lanes makes it difficult for vehicle's automatic speed control strategy to make judgments and effective control decisions. In this paper, an intelligent speed control strategy for uncertain cut-in scenarios is established based on a basic autonomous driving system. This strategy judges cut-in maneuver from surrounding vehicles and outputs adaptive control action under current environment according to Q value of state-action pair based on a Q network. In addition, according to the analysis of cut-in scenarios, the Q network is trained based on a novel reinforcement learning method named as experience screening deep Q-learning network (ES-DQN). The proposed ES-DQN is an extension of double deep Q-learning network (DDQN) algorithm, and includes two parts: experience screening and policy learning. Based on the experience screened from the experience screening part, the proposed learning method can train an intelligent speed control strategy which has stronger adaptability and control effect in uncertain cut-in scenarios. According to simulation results, the proposed intelligent speed control strategy trained by ES-DQN has better performance under uncertain cut-in scenarios than DDQN method and traditional ACC strategy. Meanwhile, by adjusting weight value in reward function, the system can realize different control target. [ABSTRACT FROM AUTHOR]

Graph learning is a popular approach for perfor ming machine learning on graph-structured data. It has revolutionized the machine learning ability to model graph data to address downstream tasks. Its application is wide due to the availability of graph data ranging from all types of networks to information systems. Most graph learning methods assume that the graph is static and its complete structure is known during training. This limits their applicability since they cannot be applied to problems where the underlying graph grows over time and/or new tasks emerge incrementally. Such applications require a lifelong learning approach that can learn the graph continuously and accommodate new information whilst retaining previously learned knowledge. Lifelong learning methods that enable continuous learning in regular domains like images and text cannot be directly applied to continuously evolving graph data, due to its irregular structure. As a result, graph lifelong learning is gaining attention from the research community. This survey paper provides a comprehensive overview of recent advancements in graph lifelong learning, including the categorization of existing methods, and the discussions of potential applications and open research problems. [ABSTRACT FROM AUTHOR]

Large-scale labeled data are generally required to train deep neural networks in order to obtain better performance in visual feature learning from images or videos for computer vision applications. To avoid extensive cost of collecting and annotating large-scale datasets, as a subset of unsupervised learning methods, self-supervised learning methods are proposed to learn general image and video features from large-scale unlabeled data without using any human-annotated labels. This paper provides an extensive review of deep learning-based self-supervised general visual feature learning methods from images or videos. First, the motivation, general pipeline, and terminologies of this field are described. Then the common deep neural network architectures that used for self-supervised learning are summarized. Next, the schema and evaluation metrics of self-supervised learning methods are reviewed followed by the commonly used datasets for images, videos, audios, and 3D data, as well as the existing self-supervised visual feature learning methods. Finally, quantitative performance comparisons of the reviewed methods on benchmark datasets are summarized and discussed for both image and video feature learning. At last, this paper is concluded and lists a set of promising future directions for self-supervised visual feature learning. [ABSTRACT FROM AUTHOR]

This papers in this special issue mainly focus on deep learning models for industry informatics, addressing both original algorithmic development and new applications of deep learning. [ABSTRACT FROM AUTHOR]

An autonomous framework to evaluate layout of a piping design in the form of piping and instrumentation diagram (P&ID) according to a set of standards of marine and offshore industry is proposed. The method starts with transforming a P&ID into a vector x in R^d . Transformation is done based on a concept introduced for piping known as Histogram of Connectivity. The proposed descriptor captures two essential properties of P&ID: attributes of each component and connectivity among the components. Next, linear support vector machine (SVM) is used to learn a classifier from existing compliant and noncompliant designs. Subsequently, the linear classifier can be used to check if an unseen design complies with the standards. In addition, to enable follow up on noncompliant design including correction or modification, a method to analyze the reason of noncompliance prediction by the learned SVM model is introduced. The method has demonstrated encouraging performance in two challenging data sets of designs created with advice from experienced engineers in the industry, based on International Convention for the Prevention of Pollution from Ships (MARPOL) and Rules for Classification of Ships of Lloyd’s Register. Note to Practitioners—This paper is motivated by need of marine and offshore industry for automated solution for design appraisal. This paper aims to address this issue by using a machine learning-based approach. Some compliant and noncompliant designs are provided to a developed algorithm for a machine (or computer) to learn. After learning is completed, the machine is able to classify unseen designs as compliant or noncompliant. As highlighted in this paper, the developed method has demonstrated encouraging performance in two case studies, including specific parts in MARPOL and Rules of Lloyd’s Register. For adoption by industry, necessary steps include collecting some designs (compliant and noncompliant) available in an organization and feeding these into the developed method for learning by machine before it can predict. With ability of highlighting possible connections that cause noncompliance, follow up and correction on a noncompliant design is made possible. [ABSTRACT FROM PUBLISHER]

Traffic flow prediction plays a critical role in reducing traffic congestion in transportation systems. However, accurate traffic flow prediction becomes challenging due to the impact of complex spatio-temporal (ST) correlations and the diversity of ST correlations. When modeling complicated ST correlations, researchers usu did not take the diversity of ST correlations into consideration, resulting in poor prediction accuracy. In this paper, we propose ST-InNet, a deep spatio-temporal Inception network for collectively predicting traffic flow in each city region. Specifically, ST-InNet employs two Inception networks to simultaneously capture various spatial and temporal correlations of traffic data, including temporal closeness, temporal periodicity, nearby spatial dependencies, and distant spatial dependencies. For the diversity of spatial correlations, ST-InNet presents an improved variant of an Inception module to explicitly capture the different contributions of spatial correlations for each region. For the diversity of temporal correlations, ST-InNet designs a fusion component to explicitly model the varying contributions of temporal correlations on prediction. The experiments are conducted on a real-world traffic dataset in Nanjing, demonstrating that ST-InNet outperforms five state-of-the-art baselines in short-term and long-term traffic flow predictions with an average accuracy improvement of 32.09% and 30.97%, respectively. [ABSTRACT FROM AUTHOR]

In this paper, we reveal that metric learning would suffer from serious inseparable problem if without informative sample mining. Since the inseparable samples are often mixed with hard samples, current informative sample mining strategies used to deal with inseparable problem may bring up some side-effects, such as instability of objective function, etc. To alleviate this problem, we propose a novel distance metric learning algorithm, named adaptive neighborhood metric learning (ANML). In ANML, we design two thresholds to adaptively identify the inseparable similar and dissimilar samples in the training procedure, thus inseparable sample removing and metric parameter learning are implemented in the same procedure. Due to the non-continuity of the proposed ANML, we develop an ingenious function, named log-exp mean function to construct a continuous formulation to surrogate it, which can be efficiently solved by the gradient descent method. Similar to Triplet loss, ANML can be used to learn both the linear and deep embeddings. By analyzing the proposed method, we find it has some interesting properties. For example, when ANML is used to learn the linear embedding, current famous metric learning algorithms such as the large margin nearest neighbor (LMNN) and neighbourhood components analysis (NCA) are the special cases of the proposed ANML by setting the parameters different values. When it is used to learn deep features, the state-of-the-art deep metric learning algorithms such as Triplet loss, Lifted structure loss, and Multi-similarity loss become the special cases of ANML. Furthermore, the log-exp mean function proposed in our method gives a new perspective to review the deep metric learning methods such as Prox-NCA and N-pairs loss. At last, promising experimental results demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Multi-order proximity is useful for effective network embedding. In contrast to many previous works that only consider order-level weights, this paper proposes to explore a more expressive node-level weighting mechanism to encode the diverse local structure, with a scalable and theoretically justified sampling strategy for its learning. Specifically, we start with a formal definition of multi-order proximity matrix which leads to our new multi-order objective based on Laplacian Eigenmaps and Skip-Gram. Then we instantiate the node-specific multi-order weights in the objective with the help of neighborhood size estimation, which indicates node-specific multi-order information. For objective learning, it is implicitly fulfilled with our proposed branching tree-like random walk strategy termed by BTWalk, which differs from the dominant chain-like walk in existing sampling techniques. BTWalk is designed by a synergetic combination of BFS (breadth-first search) and DFS (depth-first search), which is modulated according to the weights of the considered proximity orders. We theoretically analyze its cost-efficiency, and further propose the so-called Vec4Cross framework that incorporates joint node embedding and network alignment for two partially overlapped networks based on the seed matchings, whereby BTWalk is also adopted for embedding. Promising experimental results are obtained on real-world datasets across popular tasks. [ABSTRACT FROM AUTHOR]

Single image super-resolution (SISR) using deep convolutional neural networks (CNNs) achieves the state-of-the-art performance. Most existing SISR models mainly focus on pursuing high peak signal-to-noise ratio (PSNR) and neglect textures and details. As a result, the recovered images are often perceptually unpleasant. To address this issue, in this paper, we propose a texture and detail-preserving network (TDPN), which focuses not only on local region feature recovery but also on preserving textures and details. Specifically, the high-resolution image is recovered from its corresponding low-resolution input in two branches. First, a multi-reception field based branch is designed to let the network fully learn local region features by adaptively selecting local region features in different reception fields. Then, a texture and detail-learning branch supervised by the textures and details decomposed from the ground-truth high resolution image is proposed to provide additional textures and details for the super-resolution process to improve the perceptual quality. Finally, we introduce a gradient loss into the SISR field and define a novel hybrid loss to strengthen boundary information recovery and to avoid overly smooth boundary in the final recovered high-resolution image caused by using only the MAE loss. More importantly, the proposed method is model-agnostic, which can be applied to most off-the-shelf SISR networks. The experimental results on public datasets demonstrate the superiority of our TDPN on most state-of-the-art SISR methods in PSNR, SSIM and perceptual quality. We will share our code on https://github.com/tocaiqing/TDPN. [ABSTRACT FROM AUTHOR]

Decentralized learning has gained great popularity to improve learning efficiency and preserve data privacy. Each computing node makes equal contribution to collaboratively learn a Deep Learning model. The elimination of centralized Parameter Servers (PS) can effectively address many issues such as privacy, performance bottleneck and single-point-failure. However, how to achieve Byzantine Fault Tolerance in decentralized learning systems is rarely explored, although this problem has been extensively studied in centralized systems. In this paper, we present an in-depth study towards the Byzantine resilience of decentralized learning systems with two contributions. First, from the adversarial perspective, we theoretically illustrate that Byzantine attacks are more dangerous and feasible in decentralized learning systems: even one malicious participant can arbitrarily alter the models of other participants by sending carefully crafted updates to its neighbors. Second, from the defense perspective, we propose Ubar, a novel algorithm to enhance decentralized learning with Byzantine Fault Tolerance. Specifically, Ubar provides a Uniform Byzantine-resilient Aggregation Rule for benign nodes to select the useful parameter updates and filter out the malicious ones in each training iteration. It guarantees that each benign node in a decentralized system can train a correct model under very strong Byzantine attacks with an arbitrary number of faulty nodes. We conduct extensive experiments on standard image classification tasks and the results indicate that Ubar can effectively defeat both simple and sophisticated Byzantine attacks with higher performance efficiency than existing solutions. [ABSTRACT FROM AUTHOR]

Dynamic vehicle routing problems (DVRPs) have become a hot research topic due to their significance in logistics, although it is still very challenging for existing algorithms to solve DVRPs due to the dynamically changing customer requests during the optimization. In this paper, we propose a pairwise proximity learning-based ant colony algorithm, termed PPL-ACO, for tackling DVRPs. In PPL-ACO, a pairwise proximity learning method is suggested to predict the local visiting order of customers in the optimal route after the occurrence of changes, which is on the basis of learning from the optimal routes found before the changes occur. A radial basis function network is used to learn the local visiting order of customers based on the proximity between each pair of customer nodes, by which the optimal routes can be quickly tracked after changes occur. Experimental results on 22 popular DVRP instances show that the proposed PPL-ACO significantly outperforms four state-of-the-art approaches to DVRPs. More interestingly, the results on five large-scale DVRP instances demonstrate the superiority of the proposed PPL-ACO in solving large-scale DVPRs with up to 1000 customers. The results on a real case of Nankai Strict, Tianjin, China also verifies that the proposed PPL-ACO is more effective and efficient than the four compared approaches in solving real-world DVRPs. [ABSTRACT FROM AUTHOR]

Fuzzy relationships exist between students’ learning performance with various abilities and a test item. However, the challenges in implementing adaptive assessment agents are obtaining sufficient items, efficient and accurate computerized estimation, and a substantial feedback agent. Additionally, the agent must immediately estimate students’ ability item by item, which places a considerable burden on the server, especially for a group test. Hence, the implementation of an adaptive assessment agent is more difficult in practice. This paper proposes an agent with particle swarm optimization (PSO) based on a fuzzy markup language (FML) for students’ learning performance evaluation and educational applications, and the proposed agent is according to the response data from a conventional test and an item response theory (IRT)-based three-parameter logistic model. First, we apply a Gauss–Seidel based parameter estimation mechanism to estimate the items’ parameters according to the response data, and then to compare its results with those of an IRT-based Bayesian parameter estimation mechanism. In addition, we propose a static-IRT test assembly mechanism to assemble a form for the conventional test. The presented FML-based dynamic assessment mechanism infers the probability of making a correct response to the item for a student with various abilities. Moreover, this paper also proposes a novel PSO-based FML (PFML) learning mechanism for optimizing the parameters between items and students. Finally, we adopt a K-fold cross-validation mechanism to evaluate the performance of the proposed agent. Experimental results show that the novel PFML learning mechanism for the parameter estimation and learning optimization performs favorably. We believe the proposed PFML will be a reference for education research and pedagogy and an important colearning mechanism for future human–machine educational applications. [ABSTRACT FROM AUTHOR]

Domain adaptation manages to build an effective target classifier or regression model for unlabeled target data by utilizing the well-labeled source data but lying different distributions. Intuitively, to address domain shift problem, it is crucial to learn domain invariant features across domains, and most existing approaches have concentrated on it. However, they often do not directly constrain the learned features to be class discriminative for both source and target data, which is of vital importance for the final classification. Therefore, in this paper, we put forward a novel feature learning method for domain adaptation to construct both domain invariant and class discriminative representations, referred to as DICD. Specifically, DICD is to learn a latent feature space with important data properties preserved, which reduces the domain difference by jointly matching the marginal and class-conditional distributions of both domains, and simultaneously maximizes the inter-class dispersion and minimizes the intra-class scatter as much as possible. Experiments in this paper have demonstrated that the class discriminative properties will dramatically alleviate the cross-domain distribution inconsistency, which further boosts the classification performance. Moreover, we show that exploring both domain invariance and class discriminativeness of the learned representations can be integrated into one optimization framework, and the optimal solution can be derived effectively by solving a generalized eigen-decomposition problem. Comprehensive experiments on several visual cross-domain classification tasks verify that DICD can outperform the competitors significantly. [ABSTRACT FROM AUTHOR]

A model-based reinforcement learning (RL) method which interplays direct and indirect learning to update ${Q}$ functions is proposed. The environment is approximated by a virtual model that can predict the transition to the next state and the reward of the domain. This virtual model is used to train ${Q}$ functions to accelerate policy learning. Lookup table methods are usually used to establish such environmental models, but these methods need to collect tremendous amounts of experiences to enumerate responses of the environment. In this paper, a stochastic model learning method based on tree structures is presented. To model the transition probability, an online clustering method is applied to equip the model learning method with the abilities to evaluate the transition probability. By the virtual model, the RL method produces simulated experience in the stage of indirect learning. Since simulated transitions and backups are more usefully focused by working backward from the state-action, the pair estimated ${Q}$ value of which changes significantly, the useful one-step backups are actions that lead directly into the one state whose value has already obviously been changed. This, however, may induce a false positive; that is, a backup state may be an invalid state, such as an absorbing or terminal state, especially in cases where the changes of ${Q}$ values at the planning stage are still needed to put back for ranking even though they are based on a simulated experience and are possibly erroneous. It is obvious that when the agent is attracted to generate simulated experience around the area of these absorbing states, the learning efficiency is deteriorated. This paper proposes three detecting methods to solve this problem. Moreover, the policy learning can speed up. The effectiveness and generality of our method is further demonstrated in three numerical simulations. The simulation results demonstrate that the training rate of our method is obviously improved. [ABSTRACT FROM AUTHOR]

Summary: In this paper, a method is proposed to tackle the problem of single channel audio separation. The proposed method leverages on the exemplar source is used to emulate the targeted speech signal. A multicomponent nonnegative matrix factor 2D deconvolution (NMF2D) is proposed to model the temporal and spectral changes and the number of spectral basis of the audio signals. The paper proposes an artificial auxiliary channel to imitate a pair of stereo mixture signals, which is termed as “artificial‐stereophonic mixtures.” The artificial‐stereophonic mixtures and the exemplar source are jointly used to guide the factorization process of the NMF2D. The factorization is adapted under a hybrid framework that combines the generalized expectation–maximization algorithm with multiplicative update adaptation. The proposed algorithm leads to fast and stable convergence and ensures the nonnegativity constraints of the solution are satisfied. Adaptive sparsity has also been introduced on each sparse parameter in the multicomponent NMF2D model when the exemplar deviates from the target signal. Experimental results have shown the competence of the proposed algorithms in comparison with other algorithms. [ABSTRACT FROM AUTHOR]

Imbalanced training is a challenge in the field of autonomous driving. For the steering estimation task, imbalanced training is the core reason that an end-to-end model cannot estimate sharp steering value well. Inspired by researches on the steering estimation, this paper proposes a novel loss function to train a high-performance end-to-end model for handling the imbalanced training problem, which is named SteeringLoss. Firstly, the imbalanced distribution of the steering value for driving datasets is analyzed, which is similar to the double long-tailed distribution. Secondly, with the feature of distribution, this paper designs a cost-sensitive loss function step by step, the new loss function can improve the impact of the sharp steering value while maintaining the impact of the small steering value. Thirdly, three typical steering estimation models are established with SteeringLoss for demonstration, including the CNN model, the CNN-LSTM model and the 3DCNN-LSTM model. Finally, three experiments are designed for SteeringLoss: Experiment I demonstrates SteeringLoss can avoid imbalanced training problem; Experiment II discusses the finetuning principle of SteeringLoss and gives the basic guideline for using SteeringLoss; Experiment III shows the results on different typical end-to-end steering estimation models, which shows effectiveness of SteeringLoss for all the models and gives different solutions for the steering estimation. Moreover, the SteeringLoss is suitable for the imbalanced training with similar distributions of datasets besides end-to-end steering estimation, which indicates the potential value of the research on SteeringLoss in the future. [ABSTRACT FROM AUTHOR]

There has been a resurgence of research in the design of deep architecture models and learning algorithms, i.e., methods that rely on the extraction of a multilayer representation of the data. Often referred to as deep learning, this topic of research has been building on and contributing to many different research topics, such as neural networks, graphical models, feature learning, unsupervised learning, optimization, pattern recognition, and signal processing. Deep learning is also motivated and inspired by neuroscience and has had a tremendous impact on various applications such as computer vision, speech recognition, and natural language processing. The clearly multidisciplinary nature of deep learning led to a call for papers for a special issue dedicated to learning deep architectures. [ABSTRACT FROM PUBLISHER]