Search

Knowledge Graph (KG) is identified as a major area in artificial intelligence, which is used for many real-world applications. The task of knowledge graph reasoning has been widely used and proven to be effective, which aims to find these reasonable paths for various relations to solve the issue of incompleteness in KGs. However, many previous works on KG reasoning, such as path-based or reinforcement learning-based methods, are too reliant on the pre-training, where the paths from the head entity and the target entity must be given to pre-train the model, which would easily lead the model to overfit on the given paths seen in the pre-training. To address this issue, we propose a novel reasoning model named MemoryPath with a deep reinforcement learning framework, which incorporates Long Short Term Memory (LSTM) and graph attention mechanism to form the memory component. The well-designed memory component can get rid of the pre-training so that the model doesn’t depend on the given target entity for training. A tailored mechanism of reinforcement learning is presented in this proposed deep reinforcement framework to optimize the training procedure, where two metrics, Mean Selection Rate (MSR) and Mean Alternative Rate (MAR), are defined to quantitatively measure the complexities of the query relations. Meanwhile, three different training mechanisms, Action Dropout, Reward Shaping and Force Forward, are proposed to optimize the training process of the proposed MemoryPath. The proposed MemoryPath is validated on two datasets from FB15K-237 and NELL-995 on different tasks including fact prediction, link prediction and success rate in finding paths. The experimental results demonstrate that the tailored mechanism of reinforcement learning make the MemoryPath achieves state-of-the-art performance comparing with the other models. Also, the qualitative analysis indicates that the MemoryPath can store the learning process and automatically find the promising paths for a reasoning task during the training, and shows the effectiveness of the memory component.

Thyroid nodule segmentation in ultrasound images is a valuable and challenging task, and it is of great significance for the diagnosis of thyroid cancer. Due to the lack of the prior knowledge of thyroid region perception, the inherent low contrast of ultrasound images and the complex appearance changes between different frames of ultrasound video, existing automatic segmentation algorithms for thyroid nodules that directly apply semantic segmentation techniques can easily mistake non-thyroid areas as nodules. In this work, we propose a thyroid region prior guided feature enhancement network (TRFE-Net) for thyroid nodule segmentation. In order to facilitate the development of thyroid nodule segmentation, we have contributed TN3k: an open-access dataset of thyroid nodule images with high-quality nodule masks labeling. Our proposed method is evaluated on TN3k and shows outstanding performance compared with existing state-of the-art algorithms. Source code and data are available1.

China's fallow compensation policy is known to be relatively simple and unsophisticated, and in much need of improvement. In seeking such improvement, this paper designed some alternative compensation schemes that include recultivation insurance, priority right of participation in the fallowing work or free agricultural inputs, but lower income subsidy, and reported on a choice experiment of alternatives with 421 peasant-farming households in Chaling County, Hunan Province – a fallow area of heavy-metal-contaminated farmland. A mixed logit model was used to compare the choices made between pilot and non-pilot villages, and the influencing factors involved. The results showed that most farmers preferred the alternative compensation schemes to the current policy, and income subsidy was not the only factor affecting their choices, which indicated recultivation insurance and a longer period of free agricultural inputs were also important. However, the preferences differed between pilot and non-pilot village farmers, in which the pilot village farmers were more willing to accept less income subsidy in favor of increased recultivation insurance and an extended period of free agricultural inputs. By contrast, non-pilot village farmers preferred to adopt an alternative compensation scheme with a priority right of participation in the fallowing work and high level of income subsidy. The possible reason for the differences was that the publicity and training activities provided by the government had changed the preferences of pilot village farmers. Generally, the effects of various forms of economic compensation and single cash payment were different about fallow compensation policy, and the preferences of farmers from different backgrounds might also vary from person to person. Different regions needed to design the fallow compensation policy according to practical problems in the use of farmland and farmers' preferences.

Inevitable noises and limited computational time are major issues for time-variant matrix inversion in practice. When designing a time-variant matrix inversion algorithm, it is highly demanded to suppress noises without violating the performance of real-time computation. However, most existing algorithms only consider a nominal system in the absence of noises, and may suffer from a great computational error when noises are taken into account. Some other algorithms assume that denoising has been conducted before computation, which may consume extra time and may not be suitable in practice. By considering the above situation, in this paper, an enhanced discrete-time Zhang neural network (EDTZNN) model is proposed, analyzed and investigated for time-variant matrix inversion. For comparison, an original discrete-time Zhang neural network (ODTZNN) model is presented. Note that the EDTZNN model is superior to ODTZNN model in suppressing various kinds of bias noises. Moreover, theoretical analyses show the convergence of the proposed EDTZNN model in the presence of various kinds of bias noises. In addition, numerical experiments including an application to robot motion planning are provided to substantiate the efficacy and superiority of the proposed EDTZNN model for time-variant matrix inversion.

Knowledge Graph (KG) reasoning aims at finding reasoning paths for relations, in order to solve the problem of incompleteness in KG. Many previous path-based methods like PRA and DeepPath suffer from lacking memory components, or stuck in training. Therefore, their performances always rely on well-pretraining. In this paper, we present a deep reinforcement learning based model named by AttnPath, which incorporates LSTM and Graph Attention Mechanism as the memory components. We define two metrics, Mean Selection Rate (MSR) and Mean Replacement Rate (MRR), to quantitatively measure how difficult it is to learn the query relations, and take advantages of them to fine-tune the model under the framework of reinforcement learning. Meanwhile, a novel mechanism of reinforcement learning is proposed by forcing an agent to walk forward every step to avoid the agent stalling at the same entity node constantly. Based on this operation, the proposed model not only can get rid of the pretraining process, but also achieves state-of-the-art performance comparing with the other models. We test our model on FB15K-237 and NELL-995 datasets with different tasks. Extensive experiments show that our model is effective and competitive with many current state-of-the-art methods, and also performs well in practice.

Being two famous neural networks, the error back-propagation (BP) algorithm based neural networks (i.e., BP-type neural networks, BPNNs) and Hopfield-type neural networks (HNNs) have been proposed, developed, and investigated extensively for scientific research and engineering applications. They are different from each other in a great deal, in terms of network architecture, physical meaning and training pattern. In this paper of literature-review type, we present in a relatively complete and creative manner the common natures of learning between BP-type and Hopfield-type neural networks for solving various (mathematical) problems. Specifically, comparing the BPNN with the HNN for the same problem-solving task, e.g., matrix inversion as well as function approximation, we show that the BPNN weight-updating formula and the HNN state-transition equation turn out to be essentially the same. Such interesting phenomena promise that, given a neural-network model for a specific problem solving, its potential dual neural-network model can thus be developed.

In halftone image data hiding, pixel pairs containing master pixels and slave pixels are common operating units. In most previous researches, master pixels are selected at a set of pseudo-random locations, which degrade the image quality. In this paper, a secure halftone image steganographic scheme based on pair swapping is presented, which aims at minimizing the embedding distortions. Different from most previous researches, there is no master-slave relationship in the proposed scheme and the steganographic performance depends on the selection of pixel pairs instead of slave pixels. Based on a human visual system (HVS) model of halftone images, the superiority of pair swapping is proved and vertical swapping is further demonstrated to be the optimal way to improve visual quality among all pair swapping strategies. Finally, a statistical model is developed to predict the vertical pair pattern by considering its neighboring region, based on which, a distortion measurement is proposed to evaluate the embedding distortions on both vision and statistics. To play the advantage of the distortion measurement, syndrome-trellis code (STC) is employed to minimize the embedding distortions. Experimental results show that the proposed steganographic scheme achieves high statistical security with high embedding capacity without degrading the visual quality.

An earthquake-forecasting attempt is presented in this work via combining the weights and structure policy (WASP) and addition-subtraction frequency (ASF) algorithms. Specifically, based on the application of three-layer feedforward neuronets equipped with WASP algorithm, further using ASF algorithm, this work attempts to forecast a Japan earthquake with Mj 7.2 or above. Note that past earthquake dates are the only data used in this study. The feasibility and effectiveness of this attempt are verified via the numerical experiments with consistency analysis on obtained dates. Besides, according to the experimental results, an earthquake with Mj 7.2 or above may occur in August 2016 in Japan. Furthermore, with the highest possibility, the date of such an earthquake may occur is August 9, 2016.

Optical access networks, metro networks and mobile data networks are facing rapidly evolving demands, not only is it essential to satisfy the unyielding need for increased user bandwidths, but future networks must also support the growing wide variation in traffic dynamics and characteristics, due to various emerging technologies, such as cloud-based services, the Internet-of-Things (IoT) and 5G mobile systems, and due to growing trends, such as the proliferation of mobile devices and the rapidly increasing popularity of video-on-demand services. To be cost-effective and commercially sustainable, future optical networks must offer features, such as, dynamic reconfigurability, highly efficient use of network resources, elastic bandwidth provisioning with fine granularity, network sliceabilty and software defined networking (SDN). To meet these requirements Cloud Access Networks (CANs) are proposed which require a number of flexible, adaptive and reconfigurable networking elements. By exploiting digital signal processing (DSP) we have proposed a digital orthogonal filter-based multiplexing technique to implement CANs with multiplexed, independent optical channels at the wavelength, sub-wavelength, and orthogonal sub-band levels. This paper reviews the overall CAN concept, the operating principles of the various CAN network elements and presents an overview of the research work we have undertaken in order to validate the feasibility of the proposed technologies which includes real-time DSP-based demonstrations.

In this paper, the tracking-control problem of multiple-integrator (MI) systems is considered and investigated by combining Zhang dynamics (ZD) and gradient dynamics (GD). Several novel types of Zhang-gradient (ZG) controllers are proposed for the tracking control of MI systems (e.g., triple-integrator (TI) systems). As an example, the design processes of ZG controllers for TI systems with a linear output function (LOF) and/or a nonlinear output function (NOF) are presented. Besides, the corresponding theoretical analyses are elaborately given to guarantee the convergence performance of both z3g0 controllers (ZG controllers obtained by utilizing the ZD method thrice) and z3g1 controllers (ZG controllers obtained by utilizing the ZD method thrice and the GD method once) for TI systems. Numerical simulations concerning the tracking control of MI systems with different types of output functions are further performed to substantiate the feasibility and effectiveness of ZG controllers for tracking-control problems solving. Besides, comparative simulation results of the tracking control for MI systems with NOFs (e.g., y=cos(x1), y=x12+x22) substantiate that controllers of zmg1 type can resolve the singularity problem effectively with m being the times of using the ZD method.

A 100/40 Gb/s OFDM-PON with bi-directional low cost/complexity self-coherent homodyne detection is experimental demonstrated by using centralized optical comb with 0.1 nm wavelength interval to provides both the signal-carrier and the local oscillation light waves.

A digital signal process enabled dual-drive Mach–Zehnder modulator (DD-MZM)-based spectral converter is proposed and extensively investigated to realize dynamically reconfigurable and high transparent spectral conversion. As another important innovation point of the paper, to optimize the converter performance, the optimum operation conditions of the proposed converter are deduced, statistically simulated, and experimentally verified. The optimum conditions supported-converter performances are verified by detail numerical simulations and experiments in intensity-modulation and direct-detection-based network in terms of frequency detuning range-dependent conversion efficiency, strict operation transparency for user signal characteristics, impact of parasitic components on the conversion performance, as well as the converted component waveform are almost nondistortion. It is also found that the converter has the high robustness to the input signal power, optical signal-to-noise ratio variations, extinction ratio, and driving signal frequency.

The effects of triple beat interference, ASE, electric circuit and shot noise are entirely analyzed to research receiver sensitivity in MP2P uplink of OFDM-PON and draw SNR curves to guide PHY layer design of OFDM-PON.

An upgradable broadcasting service enabled all-optical OFDM-PON architecture based on centralized comb source is proposed. Optical frequency comb generator based on a re-circulating frequency shifter for our OFDM-PON system is illustrated. The configuration employs only one light source to support 8 optical network units for downstream, upstream and broadcasting services.

The evolution of biotechnological knowledge poses some new challenges to study microbial interactions. Vector autoregressive (VAR) model was proved to be an efficient approach to infer dynamic interactions in biological systems. However, high-throughput metagenomics or 16S-rRNA sequencing data is high dimension, which means that the number of covariates is much larger than the number of observations. Reducing the dimension of data or selecting suitable covariates became a critical component VAR modelling. In this paper, we develop a graph-regularised vector autoregressive model incorporating network information to infer causal relationships among microbial entities. The method not only considers the signs of the network connections among any two covariates, but also constructs a network weighted matrix by microbial topology information. The coordinate descent algorithm for estimating model parameters improves the accuracy of prediction. The experimental results on a time series data set of human gut microbiomes indicate that the proposed approach has better performance than other VAR-based models with penalty functions.

We realized a novel structured optical frequency comb generator used two cascade IM-MZMs for OFDM-PON and other next-generation optical access network. Experimental program can generate stable 16 intervals 12.5GHz flat comb spectrum and realized simply.

We described that coherent optical communication system is a new type of optical communication that using a single-frequency coherent light as the light source, the heterodyne receiving mode in the radio technology, QPSK modulation format is mostly concerned, it has a very narrow spectral width. In the condition of same symbol rate, the system capacity is twice as DPSK modulation, it can avoid various nonlinear effects effectively. The polarization multiplexing system is applied to the coherent receiver technology because POLUX is in use of two mutually perpendicular polarization states to transmit two independent signals, and compared to single polarization, we describes its advantages, In coherent demodulation, we are in use of digital signal processing methods to achieve, and in view of the simulation, we combined the technique of the QPSK and POLUX technology, In this article, we use the way of VPI simulation to realize polarization multiplexed coherent optical QPSK signal modulation and demodulation.

This paper investigates the connectivity in one-dimensional ad hoc wireless networks with an access point. We derive the probability that a random ad-hoc wireless network has exactly m clusters by means of the methods of combinatorics and probability. The probability of connectivity, i.e. m = 1, can be obtained as a special case. Further, we explain how the transmission range of node affects the connectivity of the wireless network.

Component-based configuration management is an important supporting activity of the software development management and software quality control. Based on the research of CMMI and IEEE criterion, this paper presents the 3C (control, configuration, component) model which based on the component-based and layered architecture. As a sub-model of software quality management and software process model (Mingzhi Mao and Xiaonan Luo, 2004), it was proved to be realizable and usable. Compared to other configuration model, the 3C model is structurally layered and is to move from file-based CM schemes to a component-based scheme, and this feature makes it more practicable and repeatable. The paper presents the 3C model and its realization.

Software metrics play an important role in developing high quality software as well as developer productivity. In this paper, we analyze the relationships between object-oriented (OO) metrics attributes and software quality characteristics, combining traditional and OO metrics method, present a coherent framework model for object-oriented software metrics and their computational methods, which can metric software development from four layers: function layer, class layer, module layer and system layer.

In this paper, an object-oriented (OO) estimation model, named multi-granularity object-oriented software estimation model is presented, which was based on literature [Minghzi Mao and Li Chen] and conceived to estimate the size and effort of OO products. In particular, four levels are proposed, which are function level, class level, module level and system level. With the help of regression analysis, an empirical validation is also performed, meant to assess the accuracy and effectiveness of the proposed model. Moreover, an automated estimation tool is designed and implemented at last.