Search

Self-attention, which allows transformers to capture deep bidirectional contexts, plays a vital role in BERT-like pre-trained language models. However, the maximum likelihood pre-training objective of BERT may produce an anisotropic word embedding space, which leads to biased attention scores for high-frequency tokens, as they are very close to each other in representation space and thus have higher similarities. This bias may ultimately affect the encoding of global contextual information. To address this issue, we propose TOCOL, a TOken-Level COntrastive Learning framework for improving the contextual representation of pre-trained language models, which integrates a novel self-supervised objective to the attention mechanism to reshape the word representation space and encourages PLM to capture the global semantics of sentences. Results on the GLUE Benchmark show that TOCOL brings considerable improvement over the original BERT. Furthermore, we conduct a detailed analysis and demonstrate the robustness of our approach for low-resource scenarios. [ABSTRACT FROM AUTHOR]

Light field (LF) cameras record multiple perspectives by a sparse sampling of real scenes, and these perspectives provide complementary information. This information is beneficial to LF super-resolution (LFSR). Compared with traditional single-image super-resolution, LF can exploit parallax structure and perspective correlation among different LF views. Furthermore, the performance of existing methods are limited as they fail to deeply explore the complementary information across LF views. In this paper, we propose a novel network, called the light field complementary-view feature attention network (LF-CFANet), to improve LFSR by dynamically learning the complementary information in LF views. Specifically, we design a residual complementary-view spatial and channel attention module (RCSCAM) to effectively interact with complementary information between complementary views. Moreover, RCSCAM captures the relationships between different channels, and it is able to generate informative features for reconstructing LF images while ignoring redundant information. Then, a maximum-difference information supplementary branch (MDISB) is used to supplement information from the maximum-difference angular positions based on the geometric structure of LF images. This branch also can guide the process of reconstruction. Experimental results on both synthetic and real-world datasets demonstrate the superiority of our method. The proposed LF-CFANet has a more advanced reconstruction performance that displays faithful details with higher SR accuracy than state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Sequential recommendations play a crucial role in many real-world applications. Due to the sequential nature, reinforcement learning has been employed to iteratively produce recommendations based on an observed stream of user behavior. In this setting, a recommendation agent interacts with the environments (users) by sequentially recommending items (actions) to maximize users' overall long-term cumulative rewards. However, most reinforcement learning-based recommendation models only focus on extrinsic rewards based on user feedback, leading to sub-optimal policies if user-item interactions are sparse and fail to obtain the dynamic rewards based on the users' preferences. As a remedy, we propose a dynamic intrinsic reward signal integrated with a contrastive discriminator-augmented reinforcement learning framework. Concretely, our framework contains two modules: (1) a contrastive learning module is employed to learn the representation of item sequences; (2) an intrinsic reward learning function to imitate the user's internal dynamics. Furthermore, we combine static extrinsic reward and dynamic intrinsic reward to train a sequential recommender system based on double Q-learning. We integrate our framework with five representative sequential recommendation models. Specifically, our framework augments these recommendation models with two output layers: the supervised layer that applies cross-entropy loss to perform ranking and the other for reinforcement learning. Experimental results on two real-world datasets demonstrate that the proposed framework outperforms several sequential recommendation baselines and exploration with intrinsic reward baselines. [ABSTRACT FROM AUTHOR]

In recent years, teaching machines to ask meaningful and coherent questions has attracted considerable attention in natural language processing. Question generation has found wide applications in areas such as education (testing knowledge) and chatbots (enhancing interaction). Following previous studies on conversational question generation, we propose a pretrained, encoder–decoder model that can incorporate the semantic information from both passage and hidden conversation representations. We adopt BERT as the encoder to combine external text and dialogue history, and we design a multi-head attention-based decoder to incorporate the semantic information from both text and hidden dialogue representations into the decoding process, thereby generating coherent questions. Experiments with conversational question generation and document-grounded dialogue response generation tasks indicate that the proposed model is superior to baseline models in terms of both standard metrics and human evaluations. [ABSTRACT FROM AUTHOR]

The existing pedestrian detection algorithms are not robust in the case of noise, obstruction and illumination change. To solve the problem, we propose a pedestrian detection algorithm combining the edge features of color image with the features of Histogram of Oriented Gradient on Depth image (referred to as the HOD features). The algorithm describes overall structural features of pedestrians by using shearlet transform to extract their edge features from color images, and obtains local edge features of corresponding depth images by generating HOD features. The overall structural features and local edge features are combined to form new feature descriptors to train a SVM (support vector machine) classifier. Due to the full integration of the two types of features, the algorithm shows significant advantages in pedestrian detection in the case of interfering factors such as noise, obstruction, illumination, and similar colors. The experimental results show that the detection accuracy rate of this algorithm is 15% higher than that of other algorithms when the false-positive rate is 0.1. [ABSTRACT FROM AUTHOR]

Nowadays, conference publications have gained importance both quantitatively and qualitatively. People are seeking ways to distinguish the quality and impact of different conferences. Some bibliometrics like conference impact factor have been proposed to assess them. Meanwhile, associations in some countries have implemented several projects to build conference ranking systems that classify conferences based on certain quality measures. The Computing Research and Education Association of Australasia and China Computer Federation lists are two such well-known and widely used conference rating systems. They can serve as a guide when researchers need to publish their work. At the same time, they can influence researchers’ publication decisions. In this paper, we try to find out how publication patterns in different countries have been influenced by these two lists as well as by some other factors. A random-effect Negative Binominal Regression Model is used to identify the level of the impact caused by different factors. [ABSTRACT FROM AUTHOR]

The ternary transitional metal oxide NiCo2O4 is a promising anode material for sodium ion batteries due to its high theoretical capacity and superior electrical conductivity. However, its sodium storage capability is severely limited by the sluggish sodiation/desodiation reaction kinetics. Herein, NiCo2O4 double-shelled hollow spheres were synthesized via a microwave-assisted, fast solvothermal synthetic procedure in a mixture of isopropanol and glycerol, followed by annealing. Isopropanol played a vital role in the precipitation of nickel and cobalt, and the shrinkage of the glycerol quasi-emulsion under heat treatment was responsible for the formation of the double-shelled nanostructure. The as-synthesized product was tested as an anode material in a sodium ion battery, was found to exhibit a high reversible specific capacity of 511 mAh g-1 at 100 mA g-1, and deliver high capacity retention after 100 cycles. [ABSTRACT FROM AUTHOR]

In a bird strike, the bird undergoes large deformation like flows; while most part of the structure is in small deformation, the region near the impact point may experience large deformations, even fail. This paper develops a coupled shell-material point method (CSMPM) for bird strike simulation, in which the bird is modeled by the material point method (MPM) and the aircraft structure is modeled by the Belytschko--Lin--Tsay shell element. The interaction between the bird and the structure is handled by a particle-to-surface contact algorithm. The distorted and failed shell elements will be eroded if a certain criterion is reached. The proposed CSMPM takes full advantages of both the finite element method and the MPM for bird strike simulation and is validated by several numerical examples. [ABSTRACT FROM AUTHOR]

The composition and properties of raw materials have significant influences on water retention capacity of mortar, but the basic origin remains unclear. To clarify the movement of water inside mortar and the mechanism of water retention, the dynamic dewatering experiment was designed and the mathematical model describing the dewatering process of mortar was set up in this work. The model proved to be effective according to the results from dynamic dewatering experiment. The study demonstrated that capillary gradual contraction model is suitable for describing early dewatering process of fresh mortar and the parameters in the dewatering equation are relevant to the properties of raw materials, which certify that the model is of physical significance. Moreover, influencing factors such as water properties and binding materials properties were obtained and their relationship with water retention of mortar were discussed theoretically based on capillary gradual contraction model. [ABSTRACT FROM AUTHOR]

When people want to move to a new job, it is often difficult since there is too much job information available. To select an appropriate job and then submit a resume is tedious. It is particularly difficult for university students since they normally do not have any work experience and also are unfamiliar with the job market. To deal with the information overload for students during their transition into work, a job recommendation system can be very valuable. In this research, after fully investigating the pros and cons of current job recommendation systems for university students, we propose a student profiling based re-ranking framework. In this system, the students are recommended a list of potential jobs based on those who have graduated and obtained job offers over the past few years. Furthermore, recommended employers are also used as input for job recommendation result re-ranking. Our experimental study on real recruitment data over the past four years has shown this method's potential. [ABSTRACT FROM AUTHOR]

A number of bibliometric studies point out that the role of conference publications in computer science differs from that in other traditional fields. Thus, it is interesting to identify the relative status of journal and conference publications in different subfields of computer science based on the citation rates categorised by the China Computer Federation (CCF) classifications and venue types. In this research, we construct a dataset containing over 100,000 papers recommended by the CCF catalogue and their citation information. We also investigate some other factors that often influence a paper's citation rate. An experimental study shows that the relative status of journals and conferences varies greatly in different subfields of computer science, and the impact of different publication levels varies according to the citation rate. We also verify that the classification of a publication, number of authors, maximum h-index of all authors of a paper, and average number of papers published by a publication have different effects on the citation rate, although the citation rate may have a different degree of correlation with these factors. [ABSTRACT FROM AUTHOR]

In the last decades, extensive efforts have been dedicated to develop better 3D object retrieval methods. View-based methods have attracted a significant amount of attention, not only because of their state-of-the-art performance, but also they merely require some of a 3D object's 2D view images. However, most recent approaches only deal with the images' content difference without the discrepancy of view relative positions. In this paper, we propose a normal method for view segmentation, based on Markov random field (MRF) model, which consider not only the difference between the content of views but also the relative locations. Each view is obtained by projecting at certain viewpoints and angels, therefore, these locations can be applied to depict each view, with content of views. We use the MRF to implement view segmentation and choose the representative views. Finally, we present a framework based on the proposed view segmentation method for 3D object retrieval and the experimental results demonstrate that the proposed method can achieve better retrieval effectiveness than state-of-the-art methods under several standard evaluation measures. [ABSTRACT FROM AUTHOR]

Web service is one of the most fundamental technologies in implementing service oriented architecture (SOA) based applications. One essential challenge related to web service is to find suitable candidates with regard to web service consumer's requests, which is normally called web service discovery. During a web service discovery protocol, it is expected that the consumer will find it hard to distinguish which ones are more suitable in the retrieval set, thereby making selection of web services a critical task. In this paper, inspired by the idea that the service composition pattern is significant hint for service selection, a personal profiling mechanism is proposed to improve ranking and recommendation performance. Since service selection is highly dependent on the composition process, personal knowledge is accumulated from previous service composition process and shared via collaborative filtering where a set of users with similar interest will be firstly identified. Afterwards a web service re-ranking mechanism is employed for personalised recommendation. Experimental studies are conduced and analysed to demonstrate the promising potential of this research. [ABSTRACT FROM AUTHOR]

Copyright of SCIENCE CHINA Information Sciences is the property of Springer Nature and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Copyright of SCIENCE CHINA Information Sciences is the property of Springer Nature and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

In the last few years, extensive effort has been spent to develop better performed 3-D object retrieval methods. View-based methods have attracted a significant amount of attention, not only because their state-of-art performance, but also they merely require some of a 3-D object's 2-D view images. However, most recent approaches only deal with the images' primordial-extracted features and ignore their hidden relationships. Considering these latent characters, a visual-topic-model 3-D object retrieval approach is introduced in this paper. In this framework, dense scale invariant feature transform(dense-SIFT) descriptors are extracted from a set of views of each 3-D object, and all the dense-SIFT descriptors are grouped into bag-of-word features using k-means clustering. Then, the topic distribution of a 3-D object is generated via latent dirichlet allocation (LDA) given its bag-of-word features. Gibbs sampling is applied in the learning and inference processing of LDA. We conduct experiments on the Princeton Shape Benchmark (PSB) and National Taiwan University 3D model database (NTU), and the experimental results demonstrate that the proposed method can achieve better retrieval effectiveness than the state-of-the-art methods under several standard evaluation measures. [ABSTRACT FROM AUTHOR]

Context modelling involves a) characterizing a situation with related information, and b) dealing and storing the information in a computer-understandable form. It is the keystone to enable a system to possess the perception capacity and adapt its functionality properly for different situations. However, a context model focusing on the characteristics of work-based learning is not well studied by pioneering researchers. For addressing this issue, in this work we firstly analyze several existing context models to identify the essentials of context modelling, whereby a hierarchical ontology context model is proposed to characterize work-based learning. Subsequently, we present the application of the proposed model in work-based learning scenario to provide adapted learning supports to professionals. Hence, this work has significance in both theory and practice. [ABSTRACT FROM AUTHOR]

With the development of Internet, people are more likely to post and propagate opinions online. Sentiment analysis is then becoming an important challenge to understand the polarity beneath these comments. Currently a lot of approaches from natural language processing's perspective have been employed to conduct this task. The widely used ones include bag-of-words and semantic oriented analysis methods. In this research, we further investigate the structural information among words, phrases and sentences within the comments to conduct the sentiment analysis. The idea is inspired by the fact that the structural information is playing important role in identifying the overall statement's polarity. As a result a novel sentiment analysis model is proposed based on recurrent neural network, which takes the partial document as input and then the next parts to predict the sentiment label distribution rather than the next word. The proposed method learns words representation simultaneously the sentiment distribution. Experimental studies have been conducted on commonly used datasets and the results have shown its promising potential. [ABSTRACT FROM AUTHOR]

Transmission control is a challenging problem in vehicular delay-tolerant networks because the network has a high vehicle mobility which leads to a dynamic and unreliable network connectivity. In this paper, we propose a COoperative Transmission control Scheme using erasure coding (COTS) to improve the transmission efficiency. In this scheme, the cooperative message distribution algorithm uses delivery capacities to decide the transmission direction and traffic. Besides, the cooperative task scheduling algorithm sets the transfer order for messages from both sides. This algorithm uses a comprehensive parameter, the number of data blocks to send, to simplify the computation of transmission priority. After analyzing the vehicle transmission behavior under different storage conditions, we apply local deleting and creating behaviors instead of sending and receiving behaviors in specific states to decrease communication overhead. Simulation results show that COTS, compared with other well-known schemes, not only increases the delivery ratio but also decreases the delivery latency and the transmission overhead. [ABSTRACT FROM AUTHOR]

The transport control protocol (TCP) has been widely used in wired and wireless Internet applications such as FTP, email and http. Numerous congestion avoidance algorithms have been proposed to improve the performance of TCP in various scenarios, especially for high speed and wireless networks. Although different algorithms may achieve different performance improvements under different network conditions, designing a congestion algorithm that can perform well across a wide spectrum of network conditions remains a great challenge. Delay-based TCP has a potential to overcome above challenges. However, the unfairness problem of delay-based TCP with TCP Reno blocks widely the deployment of delay-based TCP over wide area networks. In this paper, we proposed a novel delay-based congestion control algorithm, named FAST-FIT, which could perform gracefully in both ultra high speed networks and wide area networks, as well as keep graceful fairness with widely deployed TCP Reno hosts. FAST-FIT uses queuing delay as a primary input for controlling TCP congestion window. Packet loss is used as a secondary signal to adaptively adjust parameters of primary control process. Theoretical analysis and experimental results show that the performance of the algorithm is significantly improved as compared to other state-of-the-art algorithms, while maintaining good fairness. [ABSTRACT FROM AUTHOR]

Copyright of Arabian Journal for Science & Engineering (Springer Science & Business Media B.V. ) is the property of Springer Nature and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Morphology control of carbon-based nanomaterials (nanocarbons) is critical to practical applications because their physical and chemical properties are highly shape-dependent. The discovery of novel shaped nanocarbons stimulates new development in carbon science and technology. Based on direct reaction ofCO2 with Mg metal, we achieved controlled synthesis of several different types of nanocarbons including mesoporous graphene, carbon nanotubes, and hollow carbon nanoboxes. The last one, to our knowledge, has not been previously reported to this date. The method described here allows effective control of the shape and dimensions of nanocarbons through manipulation of reaction temperature. The formation mechanism of nanocarbons is proposed. As a proof of concept, the synthesized nanocarbons are used for electrodes in symmetrical supercapacitors, which exhibit high capacitance and good cycling stability. The reported protocols are instructive to production of nanocarbons with controlled shape and dimensions which are much desirable for many practical applications. [ABSTRACT FROM AUTHOR]

This paper was aimed to study the influence of modification of biochar on the performance of CO 2 adsorption. Biochar, obtained from cotton stalk pyrolysis in a fixed bed reactor, was modified with ammonia and CO 2. The physicochemical properties of biochars were characterized by the Fourier transform infrared spectroscopy and automatic adsorption equipment (Micromeritics, ASAP 2020, USA). CO 2 adsorption of biochar was performed in thermogravimetric analyzer. The results showed that the surface area of char was increased significantly by CO 2 modification, while N-contained compound on char surface was enriched obviously by NH 3 modification. CO 2 adsorption of biochar increased greatly with CO 2 and NH 3 modification. CO 2 adsorption was mainly attributed to physical adsorption at 20 °C, and the adsorption quantity (maximum = 99 mg/g) was proportional to the micropore volume of the char. However, at 120 °C, molecular thermal motion increase, chemical adsorption start to play a dominated role, and the adsorption was directly proportional to the N content of this char. [ABSTRACT FROM AUTHOR]