Search

Convolutional neural networks-based object detection techniques have achieved positive performances. However, due to the limitations of local receptive field, some existing object detection methods cannot effectively capture global information in feature extraction phases, and thus lead to unsatisfactory detection performance. Moreover, the feature information extracted by the backbone network may be redundant. To alleviate these problems, in this paper we propose a refined feature enhancement network (RFENet) for object detection. Specifically, we first propose a feature enhancement module (FEM) to capture more global and local information from feature maps with certain long-range dependencies. We further propose a multi-branch dilated attention mechanism (MDAM) to refine the extracted features in a weighted form, which can select more important spatial and channel information and broaden the receptive field of the network. Finally, we validate RFENet on MS-COCO2017, PASCAL VOC2012, and PASCAL VOC07+12 datasets, respectively. Compared to the baseline network, our RFENet improves by 2.4 AP on MS-COCO2017 dataset, 3.4 mAP on PASCAL VOC2012 dataset, and 2.7 mAP on PASCAL VOC07+12 dataset. Extensive experiments show that our RFENet can perform competitively on different datasets. The code is available at . [ABSTRACT FROM AUTHOR]

Next point of interest(POI) recommendation is a hot topic in the field of recommendation algorithms, which aims at recommending the suitable next locations for users. Recent research has significantly improved performance by simulating user interactions with POIs and the transitions between POIs using graph and sequence methods. However, existing models still have issues that need to be addressed. In response to the limitations of current next POI recommendation models, particularly in how to fully capture both global and local information on the user-POI interaction graph, and in alleviating the oversmoothing characteristics of graph neural networks that lead to information loss on the graph, this paper proposed a multi-coding network based on the graph Transformer model for recommending the next POI. Firstly, it jointly encoded global, local, and relative information on the user-POI interaction graph from the perspectives of position and structure. Then, the graph embeddings produced by this encoding were updated through graph Transformer network layers, which refreshed the information of nodes and edges on the graph. Finally, predictions were generated through MLP network layers. The MCGT model was empirically tested on two public datasets, Gowalla and TKY. The results show that at least a 3.79% improvement in recall and NDCG metrics on the Gowalla dataset and at least a 2.5% improvement on the TKY dataset, thus proving the reasonableness and effectiveness of MCGT. [ABSTRACT FROM AUTHOR]

Abstract Convolutional neural networks-based object detection techniques have achieved positive performances. However, due to the limitations of local receptive field, some existing object detection methods cannot effectively capture global information in feature extraction phases, and thus lead to unsatisfactory detection performance. Moreover, the feature information extracted by the backbone network may be redundant. To alleviate these problems, in this paper we propose a refined feature enhancement network (RFENet) for object detection. Specifically, we first propose a feature enhancement module (FEM) to capture more global and local information from feature maps with certain long-range dependencies. We further propose a multi-branch dilated attention mechanism (MDAM) to refine the extracted features in a weighted form, which can select more important spatial and channel information and broaden the receptive field of the network. Finally, we validate RFENet on MS-COCO2017, PASCAL VOC2012, and PASCAL VOC07+12 datasets, respectively. Compared to the baseline network, our RFENet improves by 2.4 AP on MS-COCO2017 dataset, 3.4 mAP on PASCAL VOC2012 dataset, and 2.7 mAP on PASCAL VOC07+12 dataset. Extensive experiments show that our RFENet can perform competitively on different datasets. The code is available at https://github.com/object9detection/RFENet .

Accurate and efficient bacterial detection is essential for public health and medical diagnostics. However, traditional detection methods are constrained by limited dataset size, complex bacterial morphology, and diverse detection environments, hindering their effectiveness. In this study, we present EagleEyeNet, a novel multi-scale information fusion model designed to address these challenges. EagleEyeNet leverages large models as teacher networks in a knowledge distillation framework, significantly improving detection performance. Additionally, a newly designed feature fusion architecture, integrating Transformer modules, is proposed to enable the efficient fusion of global and multi-scale features, overcoming the bottlenecks posed by Feature Pyramid Networks (FPN) structures, which in turn reduces information transmission loss between feature layers. To improve the model’s adaptability for different scenarios, we create our own QingDao Bacteria Detection (QDBD) dataset as a comprehensive evaluation benchmark for bacterial detection. Experimental results demonstrate that EagleEyeNet achieves remarkable performance improvements, with mAP50 increases of 3.1% on the QDBD dataset and 4.9% on the AGRA dataset, outperforming the State-Of-The-Art (SOTA) methods in detection accuracy. These findings underscore the transformative potential of integrating large models and deep learning for advancing bacterial detection technologies.

Active Learning (AL) is a technique being widely employed to minimize the time and labor costs in the task of annotating data. By querying and extracting the specific instances to train the model, the relevant task's performance is improved maximally within limited iterations. However, rare work was conducted to fully fuse features from different hierarchies to enhance the effectiveness of active learning. Inspired by the thought of information compensation in many famous deep learning models (such as ResNet, etc.), this work proposes a novel TextCNN-based Two ways Active Learning model (TCTWAL) to extract task-relevant texts. TextCNN takes the advantage of little hyper-parameter tuning and static vectors and achieves excellent results on various natural language processing (NLP) tasks, which are also beneficial to human-computer interaction (HCI) and the AL relevant tasks. In the process of the proposed AL model, the candidate texts are measured from both global and local features by the proposed AL framework TCTWAL depending on the modified TextCNN. Besides, the query strategy is strongly enhanced by maximum normalized log-probability (MNLP), which is sensitive to detecting the longer sentences. Additionally, the selected instances are characterized by general global information and abundant local features simultaneously. To validate the effectiveness of the proposed model, extensive experiments are conducted on three widely used text corpus, and the results are compared with with eight manual designed instance query strategies. The results show that our method outperforms the planned baselines in terms of accuracy, macro precision, macro recall, and macro F1 score. Especially, to the classification results on AG's News corpus, the improvements of the four indicators after 39 iterations are 40.50%, 45.25%, 48.91%, and 45.25%, respectively. [ABSTRACT FROM AUTHOR]

Circular RNAs (circRNAs) are a class of noncoding RNA molecules that are widely found in cells. Recent studies have revealed the significant role played by circRNAs in human health and disease treatment. Several restrictions are encountered because forecasting prospective circRNAs and medication sensitivity connections through biological research is not only time-consuming and expensive but also incredibly ineffective. Consequently, the development of a novel computational method that enhances both the efficiency and accuracy of predicting the associations between circRNAs and drug sensitivities is urgently needed. Here, we present DGATCCDA, a computational method based on deep learning, for circRNA–drug sensitivity association identification. In DGATCCDA, we first construct multimodal networks from the original feature information of circRNAs and drugs. After that, we adopt DeepWalk-aware graph attention networks to sufficiently extract feature information from the multimodal networks to obtain the embedding representation of nodes. Specifically, we combine DeepWalk and graph attention network to form DeepWalk-aware graph attention networks, which can effectively capture the global and local information of graph structures. The features extracted from the multimodal networks are fused by layer attention, and eventually, the inner product approach is used to construct the association matrix of circRNAs and drugs for prediction. The ultimate experimental results obtained under 5-fold cross-validation settings show that the average area under the receiver operating characteristic curve value of DGATCCDA reaches 91.18%, which is better than those of the five current state-of-the-art calculation methods. We further guide a case study, and the excellent obtained results also show that DGATCCDA is an effective computational method for exploring latent circRNA–drug sensitivity associations. [ABSTRACT FROM AUTHOR]

Effective molecular representation learning is very important for Artificial Intelligence-driven Drug Design because it affects the accuracy and efficiency of molecular property prediction and other molecular modeling relevant tasks. However, previous molecular representation learning studies often suffer from limitations, such as over-reliance on a single molecular representation, failure to fully capture both local and global information in molecular structure, and ineffective integration of multiscale features from different molecular representations. These limitations restrict the complete and accurate representation of molecular structure and properties, ultimately impacting the accuracy of predicting molecular properties. To this end, we propose a novel multi-view molecular representation learning method called MvMRL, which can incorporate feature information from multiple molecular representations and capture both local and global information from different views well, thus improving molecular property prediction. Specifically, MvMRL consists of four parts: a multiscale CNN-SE Simplified Molecular Input Line Entry System (SMILES) learning component and a multiscale Graph Neural Network encoder to extract local feature information and global feature information from the SMILES view and the molecular graph view, respectively; a Multi-Layer Perceptron network to capture complex non-linear relationship features from the molecular fingerprint view; and a dual cross-attention component to fuse feature information on the multi-views deeply for predicting molecular properties. We evaluate the performance of MvMRL on 11 benchmark datasets, and experimental results show that MvMRL outperforms state-of-the-art methods, indicating its rationality and effectiveness in molecular property prediction. The source code of MvMRL was released in https://github.com/jedison-github/MvMRL. [ABSTRACT FROM AUTHOR]

Currently, an increasing number of people are opting to use transformer models or conformer models for language identification, achieving outstanding results. Among them, transformer models based on self-attention can only capture global information, lacking finer local details. There are also approaches that employ conformer models by concatenating convolutional neural networks and transformers to capture both local and global information. However, this static single-branch architecture is difficult to interpret and modify, and it incurs greater inference difficulty and computational costs compared to dual-branch models. Therefore, in this paper, we propose a novel model called Branch-transformer (B-transformer). In contrast to traditional transformers, it consists of parallel dual-branch structures. One branch utilizes self-attention to capture global information, while the other employs a Convolutional Gated Multi-Layer Perceptron (cgMLP) module to extract local information. We also investigate various fusion methods for integrating global and local information and experimentally validate the effectiveness of our approach on the NIST LRE 2017 dataset. [ABSTRACT FROM AUTHOR]

In e-commerce, the session-based personalized recommendation remains challenging due to the limited user information within a single session. Merely relying on a user's local data is insufficient. It is vital to consider global data, extracting insights from sessions across all users to glean collaborative information. However, using all session information will waste computing resources. Moreover, much of the global data may not be pertinent to the current user, thereby undermining the quality of recommendations. To address this, we introduce the concept of personalized next session (PNS), selectively referencing sessions most relevant to the user to enhance the limited local data. This work is the first to adopt a deep network architecture study that incorporates the concept of PNS to recommend the next item for a user in the current session. We evaluated our approach on several real-world datasets, and the results show that our model outperforms state-of-the-art recommendation methods. [ABSTRACT FROM AUTHOR]

Key frame extraction can provide evidence for traffic violation detection, which is essential to support administrative punishment. However, the existing key frame extraction methods failed to model context information in complex semantic cases, such as failing to yield to pedestrian. To address this problem, we have proposed a key frame extraction model with global information balance (GIB), an intelligent vehicle violation screenshot method based on balancing the global information of video frames. The proposed GIB extracts three screenshots from the videos of vehicles failing to yield to pedestrians at crosswalks without signals. First, the proposed GIB defines the extraction of global information based on trajectories, comprising spatial structure and motion attributes as feature factors. Then, based on semantic correlation analysis for global information, relational entity filtering is implemented to avoid the interference of non-key entities and improve the effectiveness of the features. Finally, a search and pruning policy prioritizing mutual information is designed to maximize the global information entropy among preserved nodes to ensure the optimal prediction solution in case of a large global search solution space. The policy is implemented in the key frame prediction task in the Seq2Seq model based on the attention mechanism. The results of several experiments confirm the superior performance of the proposed method compared to conventional methods in terms of the evaluation of frame-time differential, perceptual hashing, and subjective scoring. For example, the perceptual hashing values of the proposed method were 10.5% and 6.7% greater than semantic correlation extraction and image similarity extraction, respectively, which are baseline methods based on local information. [ABSTRACT FROM AUTHOR]

Emotion-Cause Pair Extraction (ECPE) aims to jointly extract emotion clauses and the corresponding cause clauses from a document, which is important for user evaluation or public opinion analysis. Existing research addresses the ECPE task through a two-step or an end-to-end approach. Although previous work shows promising performances, they suffer from two limitations: 1) they fail to take full advantage of emotion type information, which has advantages for modelling the dependencies between emotion and cause clauses from a semantic perspective; 2) they ignored the interaction between local and global information, which is important for ECPE. To address these issues, we propose an ECPE Pair Generator (ECPE-PG), with a Clause-Encoder layer, a Pre-Output layer and an Information Interaction-based Pair Generation (IIPG) Module embedded. This model first encodes clauses into vector representations through the Clause-Encoder layer and then preforms emotion clause extraction (EE), cause clause extraction (CE) and emotion type extraction (ETE), respectively, through the Pre-Output layer, on the basis of which the IIPG module analyzes the complex emotional logic of relationships between clauses and estimates the candidate pairs based on the interaction of global and local information. It should be noted that emotion type information is regarded as a crucial indication in the IIPG module to assist the identification of emotion-cause pairs. Experimental results show that our method outperforms the state-of-the-art methods on benchmark datasets.

The attention mechanism can extract salient features in images, which has been proved to be effective in improving the performance of person re-identification (Re-ID). However, most of the existing attention modules have the following two shortcomings: On the one hand, they mostly use global average pooling to generate context descriptors, without highlighting the guiding role of salient information on descriptor generation, resulting in insufficient ability of the final generated attention mask representation; On the other hand, the design of most attention modules is complicated, which greatly increases the computational cost of the model. To solve these problems, this paper proposes an attention module called self-supervised recalibration (SR) block, which introduces both global and local information through adaptive weighted fusion to generate a more refined attention mask. In particular, a special ''Squeeze-Excitation'' (SE) unit is designed in the SR block to further process the generated intermediate masks, both for nonlinearizations of the features and for constraint of the resulting computation by controlling the number of channels. Furthermore, we combine the most commonly used ResNet-50 to construct the instantiation model of the SR block, and verify its effectiveness on multiple Re-ID datasets, especially the mean Average Precision (mAP) on the Occluded-Duke dataset exceeds the state-of-the-art (SOTA) algorithm by 4.49%.

In this paper, we present a unified framework to study a variety of two-person dynamic decision problems, including stochastic (zero-sum, non-zero-sum) Nash game, Stackelberg game with global information. For these games, the solvability of these problems is discussed via progressive formulations respectively: the abstract quadratic functional, Hamiltonian system for open-loop, and Riccati equation for closed-loop (feedback) representation. Based on the unified framework, time consistency/inconsistency property of related equilibrium is studied. Then we introduce a new type of game, Stackelberg game with local information. For this, the classical best-response machinery adopted for global information is no longer workable. As resolution, a repeated game approach is employed to construct the equilibrium strategies via a backward- and forward-procedure. Moreover, connection of local information pattern to time-inconsistency is also revealed. Finally, relations among zero-sum Nash game, zero-sum Stackelberg game with global information and local information are also identified. [ABSTRACT FROM AUTHOR]

Under the background of "unprecedented changes in a century", analysis of the information needs in area and country studies can provide reference for international perspectives on library and information research and practice. Through interviews with 19 scholars, the article found that the information demand in area and country studies has characteristics such as wide sources of information, multiple languages involved, and strong geographical attributes. The major problems for researchers in the process of information acquisition include insufficient and uneven distribution of domestic literature, insufficient open- ness of national libraries and archives, and scattered sources of required information. These findings can provide inspiration for innovation in concepts, platforms, models, technologies, and other aspects of information resource services from an international perspective. [ABSTRACT FROM AUTHOR]

As a combination of sound event detection and direction of arrival, the joint task of sound event localization and detection (SELD) is an emerging audio signal processing task and is applied in many areas widely. A popular convolutional recurrent neural network (CRNN)-based method uses convolution neural network (CNN) to extract high-level space features from manually designed features and utilizes recurrent neural network to model sequence context information. Some studies have shown that the normal CNN could not be robust in challenging acoustic environments such as overlapping, moving and discontinuous sources. To improve the performance of SELD in more complex acoustic scenes, parallel multi-attention enhancement (PMAE) is proposed as a convolution enhancement method to boost the representation ability of CNN in this paper. PMAE consists of attention feature enhancement (AFE) and parallel multi-attention (PMA) block. PMA, embedded into AFE, extracts boosting global–local features by efficient attention modules along with different dimensions. AFE, as a feature fusion strategy, fuses multi-scale enhanced features to improve feature representation. AFE shows great performance for overlapping sources. PMA adequately extracts characteristic information of different sound events and shows better performance on moving and discontinuous sources when it is combined with AFE. Based on such a framework, the SELD system becomes robust, while the target sources are moving and overlapping with unknown interference classes. The simulations show that proposed PMAE improves the performance enormously for SELD without other technologies, such as data augment and post-processing. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Zhejiang University (Science Edition) is the property of Journal of Zhejiang University (Science Edition) Editorial Office 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.)

Currently, an increasing number of people are opting to use transformer models or conformer models for language identification, achieving outstanding results. Among them, transformer models based on self-attention can only capture global information, lacking finer local details. There are also approaches that employ conformer models by concatenating convolutional neural networks and transformers to capture both local and global information. However, this static single-branch architecture is difficult to interpret and modify, and it incurs greater inference difficulty and computational costs compared to dual-branch models. Therefore, in this paper, we propose a novel model called Branch-transformer (B-transformer). In contrast to traditional transformers, it consists of parallel dual-branch structures. One branch utilizes self-attention to capture global information, while the other employs a Convolutional Gated Multi-Layer Perceptron (cgMLP) module to extract local information. We also investigate various fusion methods for integrating global and local information and experimentally validate the effectiveness of our approach on the NIST LRE 2017 dataset.

Optical satellite image change detection has attracted extensive research due to its comprehensive application in earth observation. Recently, deep learning (DL)-based methods have become dominant in change detection due to their outstanding performance. Remote sensing (RS) images contain different sizes of ground objects, so the information at different scales is crucial for change detection. However, the existing DL-based methods only employ summation or concatenation to aggregate several layers of features, lacking the semantic association of different layers. On the other hand, the UNet-like backbone is favored by deep learning algorithms, but the gradual downscaling and upscaling operation in the mainstream UNet-like backbone has the problem of misalignment, which further affects the accuracy of change detection. In this paper, we innovatively propose a hierarchical feature association and global correction network (HFA-GCN) for change detection. Specifically, a hierarchical feature association module is meticulously designed to model the correlation relationship among different scale features due to the redundant but complementary information among them. Moreover, a global correction module on Transformer is proposed to alleviate the feature misalignment in the UNet-like backbone, which, through feature reuse, extracts global information to reduce false alarms and missed alarms. Experiments were conducted on several publicly available databases, and the experimental results show the proposed method is superior to the existing state-of-the-art change detection models. [ABSTRACT FROM AUTHOR]

Knowledge graph completion (KGC) has been widely explored, but the task of temporal knowledge graph completion (TKGC) for predicting future events is far from perfection. Some embedding-based approaches have achieved significant results on the TKGC task by modeling the structural information of each temporal snapshot and the evolution between temporal snapshots. However, due to the uneven distribution of data in knowledge graphs (KGs), models that only utilize local structure and time series information suffer from information sparsity, resulting in some entities failing to obtain a better embedding representation due to less available information. Moreover, existing methods usually do not distinguish between the time span and frequency of historical information, which reduces the performance of link prediction. For this reason, we propose the G lobal and L ocal Information-A ware Net work (GL-ANet) to capture both global and local information. In particular, to model global information, we capture global structural information of entities across time using a global neighborhood aggregator to enrich the representation of entities; global historical information is obtained based on the frequency and time span of historical facts, focusing on recent and frequent events rather than all historical events to suggest the performance of link prediction; to model local information, we propose a two-layer attention network to capture local structural information at each timestamp, using a gating mechanism and GRU to capture local evolution information. Extensive experiments demonstrate the effectiveness of our model, achieving significant improvements and outperforming state-of-the-art models on five benchmark datasets. [ABSTRACT FROM AUTHOR]

Extractive automatic text summarization aims to extract the sentences that can best express the semantics of the full text from the original text to form a summary.It is widely used and studied due to its simplicity and efficiency.Currently,extractive summarization models are mostly based on the local relationship between sentences to obtain importance scores to select sentences.This method ignores the global semantic information of the original text,and the model is more susceptible to the influence of local non-important relationships.Therefore,an extractive summarization model incorporating global semantic information is proposed.After obtaining the representation of sentences and articles,the model learns the relationship between sentences and global information through the sentence-level encoder and global information extraction module and then integrates the extracted global information into the sentence vector.Finally,the sentence score is obtained to determine whether it is a summary sentence.The proposed model can achieve end-to-end training,and two global information extraction techniques based on aspect extraction and neural topic model are studied in the global information extraction module.Experimental results on the public dataset CNN/DailyMail verify the effectiveness of the model integrating global information.

GCN-based collaborative filtering models generate the representation of user nodes and item nodes by aggregating information on user-item interaction bipartite graph,and then predict users' preferences on items.However,they neglect users' different interaction intents and cannot fully explore the relationship between users and items.Existing graph disentangled collaborative filtering models model users' interaction intents,but ignore the global information of interaction graph and the essential features of users and items,causing the incompleteness of representation semantics.Furthermore,disentangled representation learning is inefficient due to the iterative structure of model.To solve these problems,this paper devises a deep disentangled collaborative filtering model incorporating graph global information,which is named as global graph disentangled collaborative filtering(G2DCF).G2DCF builds graph global channel and graph disentangled channel,which learns essential features and intent features,respectively.Meanwhile,by introducing orthogonality constraint and representation independence constraint,G2DCF makes every user-item interaction intent as unique as possible to prevent intent degradation,and raises the independence of representations under different intents,so as to improve the disentanglement effect.Compared with the previous graph collaborative filtering models,G2DCF can more comprehensively describe features of users and items.A number of experiments are conducted on three public datasets,and results show that the proposed method outperforms the comparison methods on multiple metrics.Further,this paper analyzes the representation distributions from independence and uniformity,verifies the disentanglement effect.It also compares the convergence speed to verify the effectiveness.

Since hyperspectral images contain a variety of ground objects of different scales, long-distance ground objects can fully extract the global spatial information of the image. However, most existing methods struggle to capture multi-scale information and global features simultaneously. Therefore, we combine two algorithms, MCNN and LSTM, and propose the MCNN–LSTM algorithm. The MCNN–LSTM model first performs multiple convolution operations on the image, and the result of each pooling layer is subjected to a feature fusion of the fully connected layer. Then, the results of fully connected layers at multiple scales and an attention mechanism are fused to alleviate the information redundancy of the network. Next, the results obtained by the fully connected layer are fed into the LSTM neural network, which enables the global information of the image to be captured more efficiently. In addition, to make the model meet the expected standard, a layer of loop control module is added to the fully connected layer of the LSTM network to share the weight information of multiple pieces of training. Finally, multiple public datasets are adopted for testing. The experimental results demonstrate that the proposed MCNN–LSTM model effectively extracts multi-scale features and global information of hyperspectral images, thus achieving higher classification accuracy. [ABSTRACT FROM AUTHOR]

The goal of dynamic scene deblurring is to remove the motion blur presented in a given image. To recover the details from the severe blurs, conventional convolutional neural networks (CNNs) based methods typically increase the number of convolution layers, kernel-size, or different scale images to enlarge the receptive field. However, these methods neglect the non-uniform nature of blurs, and cannot extract varied local and global information. Unlike the CNNs-based methods, we propose a Transformer-based model for image deblurring, named SharpFormer, that directly learns long-range dependencies via a novel Transformer module to overcome large blur variations. Transformer is good at learning global information but is poor at capturing local information. To overcome this issue, we design a novel Locality preserving Transformer (LTransformer) block to integrate sufficient local information into global features. In addition, to effectively apply LTransformer to the medium-resolution features, a hybrid block is introduced to capture intermediate mixed features. Furthermore, we use a dynamic convolution (DyConv) block, which aggregates multiple parallel convolution kernels to handle the non-uniform blur of inputs. We leverage a powerful two-stage attentive framework composed of the above blocks to learn the global, hybrid, and local features effectively. Extensive experiments on the GoPro and REDS datasets show that the proposed SharpFormer performs favourably against the state-of-the-art methods in blurred image restoration. [ABSTRACT FROM AUTHOR]

Knowing how to effectively predict the scale of future information cascades based on the historical trajectory of information dissemination has become an important topic. It is significant for public opinion guidance; advertising; and hotspot recommendation. Deep learning technology has become a research hotspot in popularity prediction, but for complex social platform data, existing methods are challenging to utilize cascade information effectively. This paper proposes a novel end-to-end deep learning network CAC-G with cascade attention convolution (CAC). This model can stress the global information when learning node information and reducing errors caused by information loss. Moreover, a novel Dynamic routing-AT aggregation method is investigated and applied to aggregate node information to generate a representation of cascade snapshots. Then, the gated recurrent unit (GRU) is employed to learn temporal information. This study's validity and generalization ability are verified in the experiments by applying CAC-G on two public datasets where CAC-G is better than the existing baseline methods. [ABSTRACT FROM AUTHOR]

Molecular property prediction is an essential but challenging task in drug discovery. The recurrent neural network (RNN) and Transformer are the mainstream methods for sequence modeling, and both have been successfully applied independently for molecular property prediction. As the local information and global information of molecules are very important for molecular properties, we aim to integrate the bi-directional gated recurrent unit (BiGRU) into the original Transformer encoder, together with self-attention to better capture local and global molecular information simultaneously. To this end, we propose the TranGRU approach, which encodes the local and global information of molecules by using the BiGRU and self-attention, respectively. Then, we use a gate mechanism to reasonably fuse the two molecular representations. In this way, we enhance the ability of the proposed model to encode both local and global molecular information. Compared to the baselines and state-of-the-art methods when treating each task as a single-task classification on Tox21, the proposed approach outperforms the baselines on 9 out of 12 tasks and state-of-the-art methods on 5 out of 12 tasks. TranGRU also obtains the best ROC-AUC scores on BBBP, FDA, LogP, and Tox21 (multitask classification) and has a comparable performance on ToxCast, BACE, and ecoli. On the whole, TranGRU achieves better performance for molecular property prediction. The source code is available in GitHub: https://github.com/Jiangjing0122/TranGRU. [ABSTRACT FROM AUTHOR]

Graph matching is a fundamental NP-problem in computer graphics and computer vision. In this work, we present an approximate graph matching method. Given two graphs to be matched, the proposed method first constructs an association graph to convert the problem of graph matching into a problem of selecting nodes on the constructed graph. The nodes of the association graph represent candidate correspondences between the two original graphs. An affinity matrix is then computed based on the local, intermediate and global information of the original graphs' nodes, each element of which is used to measure the mutual consistency of a correspondence pair within the association graph. Updating the affinity of each correspondence pair with the affinities of relevant correspondences, our method then utilizes the reweighted random walks strategy to simulate random walks on the association graph and to iteratively obtain a quasi-stationary distribution. Finally, our method applies the Hungarian algorithm to discretize the distribution. Experimental results on four common datasets verify the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Copyright of Journal of South-Central Minzu University (Natural Science Edition) is the property of Journal of South-Central Minzu University (Natural Science Edition) Editorial Office 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.)

Collaborative saliency (cosaliency) detection aims to identify common and saliency objects or regions in a set of related images. The major challenge to address is how to extract useful information on single images and image groups to express collaborative saliency cues. In this paper, we propose a global attention network for cosaliency detection to extract individual features from the feature enhancement module (FEM). Then to capture useful global information, the global information module (GIM) is applied to all individual features to obtain individual cues, and finally, group collaborative cues are obtained by the collaboration correlation module (CCM). Specifically, the channel attention module and spatial attention module are plugged into the convolution feature network. To increase global context information, we perform global information module (GIM) on the preprocessed features and embed nonlocal modules in the backbone network and adopt global average pooling to extract global semantic representation vector as individual cues. Then, we build a collaborative correlation module (CCM) to extract collaborative and consistent information by calculating the correlation between the individual features of the input image and individual cues in the collaborative correlation module. We evaluate our method on two cosaliency detection benchmark datasets (CoSal2015, iCoSeg). Extensive experiments demonstrate the effectiveness of the proposed model, in most cases our method exceeds the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Different from artistic style transfer,the challenge of photorealistic style transfer is to maintain the authenticity of the output while transferring the color style of the style input.Now,most photorealistic style transfer methods perform pre-proces-sing or post-processing based on artistic style transfer methods,to maintain the authenticity of the output image.However,artistic style transfer methods usually cannot make full use of global color information to achieve a more coordinated overall impression,and pre-processing and post-processing operations are often tedious and time-consuming.To solve the above problems,this paper establishes a photorealistic style transfer network guided by global information,and proposes a color-partition-mean loss(Lcpm) to measure the similarity of the global color distribution between output and the style input.Adaptive instance normalization(AdaIN) is improved,and partition adaptive instance normalization(AdaIN-P) is proposed to better adapt to the color style transfer of real images.In addition,this paper also introduces a cross-channel partition attention module to make better use of global context information and improve the overall coordination of output images.Through the above methods,the decoder of network is guided to make full use of global information.Experimental results show that,compared with other state-of-the-art me-thods,the proposed model can achieve a better photorealistic style transfer effect while maintaining image details.

Clustering is one of the most important techniques for unsupervised learning, it tries to divide points into different clusters without any priori knowledge of data. Therefore, several criterions for clustering algorithm are as follows: 1. Handling clusters with arbitrary shape and various density; 2. Finding cluster centers automatically; 3. Low parameter sensitivity and computational complexity. In this context, a novel algorithm namely clustering by detecting skeletal structure and identifying density fluctuation (CSSDF) was presented. In CSSDF, an efficient strategy based on density and local information of neighborhood is firstly proposed to detect the skeletal structure, which can collect the local information and identify the rough distribution of data. With the identified distribution information, a method takes expanded neighborhood and density fluctuation into consideration is proposed to further collect global information of data, which can assign all skeleton points and find cluster centers. To sum up, CSSDF can not only discover the underlying structure of data regardless of its' distribution, but also ensure the correct assignment of all skeleton points and thus lead to a satisfying clustering performance. In addition, the computational complexity of the proposed approach is O (nlogn) , which makes it possible to deal with some large clustering problem. • A novel strategy was applied to estimate density and detect skeletal structure. • Connectivity and density fluctuation were considered to assign skeletal points. • The applied two strategies contain both global and local information of data. • The algorithm is suitable for cluster with arbitrary shape and various density. [ABSTRACT FROM AUTHOR]

Text classification is the most fundamental and foundational problem in many natural language processing applications. Recently, the graph-based model (e.g., GNN-based model and GCN-based model) has been applied to this task and achieved excellent performance because of their superior capacity of modeling context from the global perspective. However, a multitude of existing graph-based models constructs a corpus-level graph structure which causes a high memory consumption and overlooks the local contextual information. To address these issues, we present a novel GNN-based model which contains a new model for building a text graph for text classification. The proposed model is called two sliding windows text GNN-based model (TSW-GNN). To be more specific, a unique text-level graph is constructed for each text, which contains a dynamic global window and a local sliding window. The local window slides inside the text to construct local word connections. Additionally, the dynamic global window slides between texts to determine word edge weights, which conquers the limitation of a single local sliding window and provides more abundant global information. We perform extensive experiments on seven benchmark datasets, and the experimental results manifest the amelioration of TSW-GNN over the most advanced models in terms of the classification accuracy. [ABSTRACT FROM AUTHOR]

Identification of RNA–small molecule binding sites plays an essential role in RNA-targeted drug discovery and development. These small molecules are expected to be leading compounds to guide the development of new types of RNA-targeted therapeutics compared with regular therapeutics targeting proteins. RNAs can provide many potential drug targets with diverse structures and functions. However, up to now, only a few methods have been proposed. Predicting RNA–small molecule binding sites still remains a big challenge. New computational model is required to better extract the features and predict RNA–small molecule binding sites more accurately. In this paper, a deep learning model, RLBind, was proposed to predict RNA–small molecule binding sites from sequence-dependent and structure-dependent properties by combining global RNA sequence channel and local neighbor nucleotides channel. To our best knowledge, this research was the first to develop a convolutional neural network for RNA–small molecule binding sites prediction. Furthermore, RLBind also can be used as a potential tool when the RNA experimental tertiary structure is not available. The experimental results show that RLBind outperforms other state-of-the-art methods in predicting binding sites. Therefore, our study demonstrates that the combination of global information for full-length sequences and local information for limited local neighbor nucleotides in RNAs can improve the model's predictive performance for binding sites prediction. All datasets and resource codes are available at https://github.com/KailiWang1/RLBind. [ABSTRACT FROM AUTHOR]

The article describes libraries, their place and activity in the global information space. Using the experience of world libraries, the role of libraries in managing document-information flow and directions of activity were reflected in the research. As a social institution, the duties of libraries and the level of their openness to innovations in the socialized information environment are determined. The importance of the promotion of multicultural values in libraries in the informationized environment is increasing day by day. The expansion of global processes due to their scale has led to the deepening of the research field of all scientific fields, as well as social and humanitarian sciences, social philosophy and cultural studies. They include various cultural concepts, philosophical and cultural theories about the future of mankind. The development and improvement of these theories leads to an understandable explanation of the development process of the new world order. Thus, in addition to joining the social and political processes to solve the global problems of the Republic of Azerbaijan, he is worried about having a national thought and thinking, preserving the national presence, and performing the function of propaganda in order to protect libraries. and the machine in the high-level development of these currents. The course of history has come a long way from the information space to the modern one. [ABSTRACT FROM AUTHOR]

In this work, we consider transferring global information from Transformer to Convolutional Neural Network (CNN) for medical semantic segmentation tasks. Previous network models for medical semantic segmentation tasks often suffer from difficulties in modeling global information or oversized model parameters. Here, to design a compact network with global and local information, we extract the global information modeling capability of Transformer into the CNN network and successfully apply it to the medical semantic segmentation tasks, called Global Information Distillation. In addition, the following two contributions are proposed to improve the effectiveness of distillation: i) We present an Information Transfer Module, which is based on a convolutional layer to prevent over-regularization and a Transformer layer to transfer global information; ii) For purpose of better transferring the teacher's soft targets, a Shrinking Result-Pixel distillation method is proposed in this paper. The effectiveness of our knowledge distillation approach is demonstrated by the experiments on multi-organ and cardiac segmentation tasks. [ABSTRACT FROM AUTHOR]

Smoke is translucent and irregular, resulting in a very complex mix between background and smoke. Thin or small smoke is visually inconspicuous, and its boundary is often blurred. Therefore, it is a very difficult task to completely segment smoke from images. To solve the above issues, a multi-scale semantic segmentation for fire smoke based on global information and U-Net is proposed. This algorithm uses multi-scale residual group attention (MRGA) combined with U-Net to extract multi-scale smoke features, and enhance the perception of small-scale smoke. The encoder Transformer was used to extract global information, and improve accuracy for thin smoke at the edge of images. Finally, the proposed algorithm was tested on smoke dataset, and achieves 91.83% mIoU. Compared with existing segmentation algorithms, mIoU is improved by 2.87%, and mPA is improved by 3.42%. Thus, it is a segmentation algorithm for fire smoke with higher accuracy. [ABSTRACT FROM AUTHOR]

This study investigated how global and local information about attentional demands influence attentional control, with a special interest in whether one information source dominates when they conflict. In Experiment 1, we manipulated proportion congruence in two blocks (i.e., mostly congruent versus mostly incongruent) of a Stroop task to create different global demands (i.e., low versus high, respectively). Additionally, we created different local demands by embedding 10-trial lists in each block that varied in their proportion congruence (10% to 90% congruent), and half the lists were preceded by a valid precue explicitly informing participants of upcoming attentional demands. Stroop effects were smaller in mostly incongruent compared with mostly congruent blocks demonstrating the influence of global information. Stroop effects also varied according to the proportion congruence of the abbreviated lists and differed between cued and uncued lists (i.e., cueing effect), demonstrating the influence of local information. Critically, we found that global and local information interacted, such that the cueing effect differed between the two blocks. While there was evidence that participants used the precue to relax control for mostly congruent lists within the mostly congruent block, the cueing effect was absent within the mostly incongruent block. In Experiment 2, we replicated the latter pattern and thereby provided further evidence that participants do not use local precues to relax control when attentional demands are globally high. The findings suggest that both global and local information sources influence the control of attention, and global information dominates local expectations when the information sources collide. [ABSTRACT FROM AUTHOR]

Since the convolutional operation pays too much attention to local information, resulting in the loss of global information and a decline in fusion quality. In order to ensure that the fused image fully captures the features of the entire scene, an end-to-end Multi-scale Global Feature Autoencoder (MGFA) is proposed in this paper, which can generate fused images with both global and local information. In this network, a multi-scale global feature extraction module is proposed, which combines dilated convolutional modules with the Global Context Block (GCBlock) to extract the global features ignored by the convolutional operation. In addition, an adaptive embedded residual fusion module is proposed to fuse different frequency components in the source images with the idea of embedded residual learning. This can enrich the detailed texture of the fused results. Extensive qualitative and quantitative experiments have demonstrated that the proposed method can achieve excellent results in retaining global information and improving visual effects. Furthermore, the fused images obtained in this paper are more adapted to the object detection task and can assist in improving the precision of detection. • An end-to-end Multi-scale Global Feature Autoencoder is proposed to fuse images. • Multi-scale global feature extraction module is proposed to extract global features. • Adaptive embedded residual fusion module is proposed by embedded residual learning. • The proposed method can achieve excellent results in improving visual effects. • The fused images obtained from MGFA are more adapted to the object detection task. [ABSTRACT FROM AUTHOR]

Keyphrase Extraction is an automatic task that involves identifying the key words or phrases that capture the main content of an article. It is useful for various downstream tasks, including text search, text clustering, and text classification. Embedding-based methods for keyphrase extraction have shown promising results by utilizing pre-trained language models to represent candidate phrases and documents separately. These methods then rank the candidate phrases based on the cosine similarity between the document and the candidate phrases embeddings. However, there are mainly two shortcomings in such methods: I) Redundancy errors, when there are partial repetitions of candidate keyphrases, the methods tend to use redundant long phrases as keyphrases; II) Low keyphrase coverage, such as some keyphrases used to describe locally important information are ignored. In this paper, we propose an unsupervised keyphrase extraction method called "MICRank", which evaluates the importance of candidate keyphrases from three perspectives: global information, local information, and attribute information, and solved the aforementioned issues. The experimental results on six benchmarks demonstrate that the proposed MICRank method outperforms the state-of-the-art unsupervised keyphrase extraction methods. In addition, this paper improves the judgment criterion of correct keyphrase extraction and introduces a new evaluation metric called S1@M (M ∈ {5,10,15}) to address the issue of synonyms being considered incorrect predictions. [ABSTRACT FROM AUTHOR]

The fusion of infrared and visible images aims to extract and fuse thermal target information and texture details to the fullest extent possible, enhancing the visual understanding capabilities of images for both humans and computers in complex scenes. However, existing methods have difficulties in preserving the comprehensiveness of source image feature information and enhancing the saliency of image texture information. Therefore, we put forward a novel infrared and visible image fusion algorithm based on global information-enhanced attention network (GIEA). Specifically, we develop an attention-guided Transformer module (AGTM) to make sure the fused images have enough global information. This module combines the convolutional neural network and Transformer to perform adequate feature extraction from shallow to deep layers, and utilize the attention network for multi-level feature-guided learning. Then, we build the contrast enhancement module (CENM), which enhances the feature representation and contrast of the image so that the fused image contains significant texture information. Furthermore, our network is driven to fully preserve the texture and structure details of the source images with a loss function that consists of content loss and total variance loss. Numerous experiments demonstrate that our fusion approach outperforms other fusion approaches in both subjective and objective assessments. • Combining CNN and Transformer to fully extract complementary features. • Designed an attention network for multi-level feature learning. • Designed a contrast enhancement module to enhance feature saliency. [ABSTRACT FROM AUTHOR]

Complex image scenes are a challenge in the collaborative saliency object detection task in the field of saliency detection, such as the inability to accurately locate salient object, surrounding background information affecting object recognition, and the inability to fuse multi-layer collaborative features well. To solve these problems, we propose a long-range dependent and dual-stream guided feature fusion network. Firstly, we enhance saliency feature by the proposed coordinate attention module so that the network can learn a better feature representation. Secondly, we capture the long-range dependency information of image feature by the proposed non-local module, to obtain more comprehensive contextual complex information. At lastly, we propose a dual-stream guided network to fuse multiple layers of synergistic saliency features. The dual-stream guided network includes classification streams and mask streams, and the layers in the decoding network are guided to fuse the feature of each layer to output more accurate synoptic saliency prediction map. The experimental results show that our method is superior to the existing methods on three common datasets: CoSal2015, CoSOD3k, and CoCA. • Our model enables the network to learn more feature representation. • Our model can capture long-distance dependent and enhance detection capability. • Our model can fuse multi-layer collaborative saliency feature. [ABSTRACT FROM AUTHOR]