Search

State-of-the-art instance segmentation approaches, such as Mask R-CNN, have exhibited remarkable performance under clear weather conditions. However, their effectiveness is significantly compromised in foggy environments, primarily due to reduced visibility and obscured object details. To address this challenge, we introduce a joint defogging learning with boundary refinement (JDLMask) framework. Unlike conventional strategies that treat image dehazing as a preprocessing step, JDLMask employs a shared structure that enables the joint learning of defogging and instance segmentation. This integrated approach greatly bolsters the model's adaptability to foggy scenarios. Recognizing challenges in feature extraction due to fog interference, we propose a multi-scale feature fusion mask head, based on the encoder–decoder architecture. This component is designed to acquire both local and global information, thereby enhancing the model's feature representation capacity. Furthermore, we integrate a boundary refinement module, which sharpens the model's localization accuracy by focusing on critical boundary details. Addressing the scarcity of datasets tailored, for instance, segmentation in real-world foggy scenes, we have enriched the Foggy Driving dataset with meticulously crafted instance mask annotations and named it the Foggy Driving InstanceSeg. Comprehensive experiments demonstrate JDLMask's superiority. Compared to the baseline Mask R-CNN, JDLMask achieves improvements of 4.4% and 3.8% in mask AP on the Foggy Cityscapes and Cityscapes validation sets, respectively, and a 2.5% gain on the Foggy Driving InstanceSeg dataset. [ABSTRACT FROM AUTHOR]

Background: Recently, magnetic resonance imaging (MRI) has become a crucial medical imaging technology widely used in clinical practice. However, MRI faces challenges such as the lengthy acquisition time for k‐space data and the need for time‐consuming manual annotation by radiologists. Traditionally, these challenges have been addressed individually through undersampled MRI reconstruction and automatic segmentation algorithms. Whether undersampled MRI segmentation can be enhanced by treating undersampled MRI reconstruction and segmentation as an end‐to‐end task, trained simultaneously, rather than as serial tasks should be explored. Purpose: We introduce a novel Transformer‐ and Joint Learning‐based Dual‐domain Network (TJLD‐Net) for undersampled MRI segmentation. Methods: This method significantly enhances feature recognition in the segmentation process by fully utilizing the rich detail obtained during the image reconstruction phase. Consequently, the method can achieve precise and reliable image segmentation even with undersampled k‐space data. Additionally, it incorporates an attention mechanism for feature enhancement, which improves the representation of shared features by learning the contextual information in MR images. Results: Simulation experiments demonstrate that the segmentation performance of TJLD‐Net on three datasets is significantly higher than that of the joint model (RecSeg) and six baseline models (where reconstruction and segmentation are regarded as serial tasks). On the CHAOS dataset, the Dice scores of TJLD‐Net are, on average, 9.87%, 2.17%, 1.90%, 1.80%, 9.60%, 0.80%, and 6.50% higher than those of the seven compared models. On the ATLAS challenge dataset, the average Dice scores of TJLD‐Net improve by 4.23%, 5.63%, 2.30%, 1.53%, 3.57%, 0.93%, and 6.60%. Similarly, on the SKM‐TEA dataset, the average Dice scores of TJLD‐Net improve by 4.73%, 12.80%, 14.83%, 8.67%, 4.53%, 11.60%, and 12.10%. The novel TJLD‐Net model provides a promising solution for undersampled MRI segmentation, overcoming the poor performance issues encountered by automated segmentation algorithms in low‐quality accelerated imaging. [ABSTRACT FROM AUTHOR]

Understanding spoken language is crucial for conversational agents, with intent detection and slot filling being the primary tasks in natural language understanding (NLU). Enhancing the NLU tasks can lead to an accurate and efficient virtual assistant thereby reducing the need for human intervention and expanding their applicability in other domains. Traditionally, these tasks have been addressed individually, but recent studies have highlighted their interconnection, suggesting better results when solved together. Recent advances in natural language processing have shown that pretrained word embeddings can enhance text representation and improve the generalization capabilities of models. However, the challenge of poor generalization in joint learning models for intent detection and slot filling remains due to limited annotated datasets. Additionally, traditional models face difficulties in capturing both the semantic and syntactic nuances of language, which are vital for accurate intent detection and slot filling. This study proposes a hybridized text representation method using a multichannel convolutional neural network with three embedding channels: non-contextual embeddings for semantic information, part-of-speech (POS) tag embeddings for syntactic features, and contextual embeddings for deeper contextual understanding. Specifically, we utilized word2vec for non-contextual embeddings, one-hot vectors for POS tags, and bidirectional encoder representations from transformers (BERT) for contextual embeddings. These embeddings are processed through a convolutional layer and a shared bidirectional long short-term memory (BiLSTM) network, followed by two softmax functions for intent detection and slot filling. Experiments on the air travel information system (ATIS) and SNIPS datasets demonstrated that our model significantly outperformed the baseline models, achieving an intent accuracy of 97.90% and slot filling F1-score of 98.86% on the ATIS dataset, and an intent accuracy of 98.88% and slot filling F1-score of 97.07% on the SNIPS dataset. These results highlight the effectiveness of our proposed approach in advancing dialogue systems, and paving the way for more accurate and efficient natural language understanding in real-world applications. [ABSTRACT FROM AUTHOR]

Recent advances in spatial transcriptomics (ST) enable measurements of transcriptome within intact biological tissues by preserving spatial information, offering biologists unprecedented opportunities to comprehensively understand tissue micro-environment, where spatial domains are basic units of tissues. Although great efforts are devoted to this issue, they still have many shortcomings, such as ignoring local information and relations of spatial domains, requiring alternatives to solve these problems. Here, a novel algorithm for spatial domain identification in S patial T ranscriptomics data with S tructure C orrelation and S elf- R epresentation (ST-SCSR), which integrates local information, global information, and similarity of spatial domains. Specifically, ST-SCSR utilzes matrix tri-factorization to simultaneously decompose expression profiles and spatial network of spots, where expressional and spatial features of spots are fused via the shared factor matrix that interpreted as similarity of spatial domains. Furthermore, ST-SCSR learns affinity graph of spots by manipulating expressional and spatial features, where local preservation and sparse constraints are employed, thereby enhancing the quality of graph. The experimental results demonstrate that ST-SCSR not only outperforms state-of-the-art algorithms in terms of accuracy, but also identifies many potential interesting patterns. [ABSTRACT FROM AUTHOR]

Abstract Advances in single-cell transcriptomics provide an unprecedented opportunity to explore complex biological processes. However, computational methods for analyzing single-cell transcriptomics still have room for improvement especially in dimension reduction, cell clustering, and cell–cell communication inference. Herein, we propose a versatile method, named DcjComm, for comprehensive analysis of single-cell transcriptomics. DcjComm detects functional modules to explore expression patterns and performs dimension reduction and clustering to discover cellular identities by the non-negative matrix factorization-based joint learning model. DcjComm then infers cell–cell communication by integrating ligand-receptor pairs, transcription factors, and target genes. DcjComm demonstrates superior performance compared to state-of-the-art methods.

In this paper, we propose a hybrid network structure to achieve more discriminant feature representations for speaker recognition, termed Resformer, local frame-level features are extracted by convolution operation, and global segment-level features are extracted by transformer, which the information learned from the frame-level features is used to guide the segment-level feature learning, to take advantage of convolutional operations and self-attention mechanisms for enhanced representation learning. Resformer uses an interactive approach to combine local features and global representations at varying resolutions. With a concurrent structure, Resformer ensures that both local features and global representations are preserved to the fullest extent possible. Extensive experiments are performed on VoxCeleb, a public SV dataset, and the findings indicate that Resformer outperforms existing deep embedding architectures by a margin of at least 10–37% on the test set. Our proposed approaches are shown to achieve significant improvement over prior methods through the ablation experiments. [ABSTRACT FROM AUTHOR]

The article addresses document-level event extraction challenges, proposing a novel framework with a token-event-role data structure and a multi-channel prediction module. Topics include the significance of document-level event extraction, challenges in multi-event detection and argument identification, and previous methods' limitations.

Understanding spoken language is crucial for conversational agents, with intent detection and slot filling being the primary tasks in natural language understanding (NLU). Enhancing the NLU tasks can lead to an accurate and efficient virtual assistant thereby reducing the need for human intervention and expanding their applicability in other domains. Traditionally, these tasks have been addressed individually, but recent studies have highlighted their interconnection, suggesting better results when solved together. Recent advances in natural language processing have shown that pretrained word embeddings can enhance text representation and improve the generalization capabilities of models. However, the challenge of poor generalization in joint learning models for intent detection and slot filling remains due to limited annotated datasets. Additionally, traditional models face difficulties in capturing both the semantic and syntactic nuances of language, which are vital for accurate intent detection and slot filling. This study proposes a hybridized text representation method using a multichannel convolutional neural network with three embedding channels: non-contextual embeddings for semantic information, part-of-speech (POS) tag embeddings for syntactic features, and contextual embeddings for deeper contextual understanding. Specifically, we utilized word2vec for non-contextual embeddings, one-hot vectors for POS tags, and bidirectional encoder representations from transformers (BERT) for contextual embeddings. These embeddings are processed through a convolutional layer and a shared bidirectional long short-term memory (BiLSTM) network, followed by two softmax functions for intent detection and slot filling. Experiments on the air travel information system (ATIS) and SNIPS datasets demonstrated that our model significantly outperformed the baseline models, achieving an intent accuracy of 97.90% and slot filling F1-score of 98.86% on the ATIS dataset, and an intent accuracy of 98.88% and slot filling F1-score of 97.07% on the SNIPS dataset. These results highlight the effectiveness of our proposed approach in advancing dialogue systems, and paving the way for more accurate and efficient natural language understanding in real-world applications.

Abstract Advances in spatial transcriptomics provide an unprecedented opportunity to reveal the structure and function of biology systems. However, current algorithms fail to address the heterogeneity and interpretability of spatial transcriptomics data. Here, we present a multi-layer network model for identifying spatial domains in spatial transcriptomics data with joint learning. We demonstrate that spatial domains can be precisely characterized and discriminated by the topological structure of cell networks, facilitating identification and interpretability of spatial domains, which outperforms state-of-the-art baselines. Furthermore, we prove that network model offers an effective and efficient strategy for integrative analysis of spatial transcriptomics data from various platforms.

Event extraction (EE) generally contains two subtasks: viz., event detection and argument extraction. Owing to the success of machine reading comprehension (MRC), some researchers formulate EE into MRC frameworks. However, existing MRC-based EE techniques are pipeline methods that suffer from error propagation. Moreover, the correlation between event types and argument roles is pre-defined by experts, which is time-consuming and inflexible. To avoid these issues, event detection and argument extraction are formalized as joint MRC. Different from previous methods, which just generate questions for argument roles for identified event types, questions are generated for all arguments that appear in the given sentence in our approach. Moreover, an end-to-end MRC model, JEEMRC, is proposed, which consists of an event classifier and a machine reader with a coarse-to-fine attention mechanism. Our proposed model can train two subtasks jointly to alleviate error propagation and utilizes interaction information between event types and argument roles to improve the performance of both tasks. Experiments on ACE 2005 verify that our JEEMRC achieves competitive results compared with previous work. In addition, it performs well when detecting events and extracting arguments in data-scarce scenarios. [ABSTRACT FROM AUTHOR]

Motivation The technology for analyzing single-cell multi-omics data has advanced rapidly and has provided comprehensive and accurate cellular information by exploring cell heterogeneity in genomics, transcriptomics, epigenomics, metabolomics and proteomics data. However, because of the high-dimensional and sparse characteristics of single-cell multi-omics data, as well as the limitations of various analysis algorithms, the clustering performance is generally poor. Matrix factorization is an unsupervised, dimensionality reduction-based method that can cluster individuals and discover related omics variables from different blocks. Here, we present a novel algorithm that performs joint dimensionality reduction learning and cell clustering analysis on single-cell multi-omics data using non-negative matrix factorization that we named scMNMF. We formulate the objective function of joint learning as a constrained optimization problem and derive the corresponding iterative formulas through alternating iterative algorithms. The major advantage of the scMNMF algorithm remains its capability to explore hidden related features among omics data. Additionally, the feature selection for dimensionality reduction and cell clustering mutually influence each other iteratively, leading to a more effective discovery of cell types. We validated the performance of the scMNMF algorithm using two simulated and five real datasets. The results show that scMNMF outperformed seven other state-of-the-art algorithms in various measurements. Availability and implementation scMNMF code can be found at https://github.com/yushanqiu/scMNMF. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Computer Engineering & Applications is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. 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.)

Previous researches on implicit discourse relationship recognition (IDRR) usually focus on designing effective discourse encoders. Different from theirs, this paper proposes a novel approach which introduces contrastive learning into IDRR so as to obtain representations of discourse units (DUs) with more differentiation. Specifically, a lightweight IDRR classification model is firstly adopted. Then, to better learn representations of DUs, the application of three different contrastive learning methods in IDDR are explored from multiple views, including instance-level, batch-level, and group-level. Finally, three multi-view contrastive learning objectives are combined for better IDRR. Our proposed method only slightly increases training time and introduces small additional parameters. Experimental results on PDTB 2. 0 show that our method achieves the state-of-the-art performance. [ABSTRACT FROM AUTHOR]

The traditional multi-task Takagi-Sugeno-Kang (TSK) fuzzy system modeling methods pay more attention to utilizing the inter-task correlation to learn the consequent parameters but ignore the importance of the antecedent parameters of the model. To this end, we propose a novel multi-task TSK fuzzy system modeling method based on multi-task fuzzy clustering. This method first proposes a novel multi-task fuzzy c-means clustering method that learns multiple specific clustering centers for each task and some common clustering centers for all tasks. Secondly, for the consequent parameters of the fuzzy system, the novel low-rank and row-sparse constraints are proposed to better implement multi-task learning. The experimental results demonstrate that the proposed model shows better performance compared with other existing methods. [ABSTRACT FROM AUTHOR]

There are many heterogeneous graphs in the real world, including reference, movie, medical, and transportation networks. In the real world, heterogeneous graphs often have multiple types of nodes and different kinds of edges. To learn node representations of heterogeneous graphs, it is necessary to incorporate rich information about target nodes into low-dimensional vector representations for downstream task analysis. However, most existing heterogeneous models only consider information from low-order views and ignore high-order node information. This will cause the target node to aggregate only local structural information while disregarding global structural information. In addition, most existing dual-view fusion methods only use the method of minimizing cosine similarity, ignoring the role of mutual information. They can respectively capture the correlation and similarity between nodes in the graph, so they are equally important. To address these limitations, we propose a low-order and high-order joint learning model (HSGNN) for heterogeneous graph networks. The model is divided into three main parts: a low-order view feature aggregation module, a high-order view feature aggregation module, and a dual-view joint learning module. The low-order view feature aggregation module obtains the low-order information of the target node through layer-by-layer aggregation. The low-level view feature aggregation module obtains local structural information of the target node through layer-by-layer aggregation. The high-order view feature aggregation module utilizes the denoised reachable matrix to obtain the global structural information of the target node. The dual view joint learning module combines cosine similarity and mutual information to learn the correlation and similarity between nodes. Finally, we conducted node classification and node clustering experiments on three real-world datasets, which improved by approximately 2 percent and 10 percent respectively, demonstrating the effectiveness of HSGNN.

Tumor samples clustering based on subspace segmentation is an effective method to discover cancer subtypes. Accurate and reliable identifications of cancer subtypes are crucial for understanding cancer pathogenesis as well as clinical diagnosis and treatment. Joint learning-based subspace clustering methods utilize the high correlation between data affinity and data segmentation for better clustering results. However, existing joint learning-based methods only provide an approximation of the segmentation cost in the joint optimization term, which could lead to sub-optimal results. To address this problem, we propose an algorithm named joint learning of spectral clustering and low-rank representation based on precise segmentation cost (JLSLPS) for cancer subtype identification. In our method, we impose non-negative and symmetric constraints on the low-rank representation matrix so that the representation coefficient can be equivalent to data affinity for the precise representation of segmentation cost. Therefore, the spectral clustering objective can be represented precisely to guide the learning of data affinity and segmentation more effectively. Finally, we solve the optimization problem of JLSLPS by using the linearized alternating direction method with adaptive penalty. We run JLSLPS on 8 cancer gene expression datasets and used 9 state-of-the-art clustering methods for comparison. Experimental results show that our method can increase ACC by 1.00-7.95% and NMI by 2.7-12.22% compared with the other methods, which proves the superiority of the proposed method.

As countries transition from external assistance while pursuing ambitious plans to achieve universal health coverage (UHC), there is increasing need to facilitate knowledge sharing and learning among them. Country-led and country-owned knowledge management is foundational to sustainable, more equitable external assistance for health and is a useful complement to more conventional capacity-building modalities provided under external assistance. In the context of external assistance, few initiatives use country-to-country sharing of practitioner experiences, and link learning to receiving guidance on how to adapt, apply and sustain policy changes. Dominant knowledge exchange processes are didactic, implicitly assuming static technical needs, and that practitioners in low- and middle-income countries require problem-specific, time-bound solutions. In reality, the technical challenges of achieving UHC and the group of policymakers involved continuously evolve. This paper aims to explore factors which are supportive of experience-based knowledge exchange between practitioners from diverse settings, drawing from the experience of the Joint Learning Network (JLN) for UHC—a global network of practitioners and policymakers sharing experiences about common challenges to develop and implement knowledge products supporting reforms for UHC—as an illustration of a peer-to-peer learning approach. This paper considers: (1) an analysis of JLN monitoring and evaluation data between 2020 and 2023 and (2) a qualitative inquiry to explore policymakers' engagement with the JLN using semi-structured interviews (n = 14) with stakeholders from 10 countries. The JLN's experience provides insights to factors that contribute to successful peer-to-peer learning approaches. JLN relies on engaging a network of practitioners with diverse experiences who organically identify and pursue a common learning agenda. Meaningful peer-to-peer learning requires dynamic, structured interactions, and alignment with windows of opportunity for implementation that enable rapid response to emerging and timely issues. Peer-to-peer learning can facilitate in-country knowledge sharing, learning and catalyse action at the institutional and health system levels. [ABSTRACT FROM AUTHOR]

With the recent advent of single-cell level biological understanding, a growing interest is in identifying cell states or subtypes that are homogeneous in terms of gene expression and are also enriched in certain biological conditions, including disease samples versus normal samples (condition-specific cell subtype). Despite the importance of identifying condition-specific cell subtypes, existing methods have the following limitations: since they train models separately between gene expression and the biological condition information, (1) they do not consider potential interactions between them, and (2) the weights from both types of information are not properly controlled. Also, (3) they do not consider non-linear relationships in the gene expression and the biological condition. To address the limitations and accurately identify such condition-specific cell subtypes, we develop scDeepJointClust, the first method that jointly trains both types of information via a deep neural network. scDeepJointClust incorporates results from the power of state-of-the-art gene-expression-based clustering methods as an input, incorporating their sophistication and accuracy. We evaluated scDeepJointClust on both simulation data in diverse scenarios and biological data of different diseases (melanoma and non-small-cell lung cancer) and showed that scDeepJointClust outperforms existing methods in terms of sensitivity and specificity. scDeepJointClust exhibits significant promise in advancing our understanding of cellular states and their implications in complex biological systems. [ABSTRACT FROM AUTHOR]

Recently, unsupervised domain adaptive person re-identification (Re-ID)methods have been extensively studied thanks to not requiring annotations, and they have achieved excellent performance. Most of the existing methods aim to train the Re-ID model for learning a discriminative feature representation. However, they usually only consider training the model to learn a global feature of a pedestrian image, but neglecting the local feature, which restricts further improvement of model performance. To address this problem, two local branches are added to the networks, aiming to allow the model to focus on the local feature containing identity information. Furthermore, we propose a self-supervised consistency constraint to further improve robustness of the model. Specifically, the self-supervised consistency constraint uses the basic data augmentation operations without other auxiliary networks, which can improve performance of the model effectively. Then, a learnable memory matrix is designed to store the mapping vectors that maps person features into probability distributions. Finally, extensive experiments are conducted on multiple commonly used person Re-ID datasets to verify the effectiveness of the proposed generative adversarial networks fusing global and local features. Experimental results reveal that our method achieves results comparable to state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Accurate and reliable segmentation of polyp targets is crucial in the treatment of colorectal cancer. However, in clinical practice, polyp structures represent only a small portion of the image, which is essential for the accurate diagnosis and treatment of colorectal cancer. The small size of lesion areas leads to a reduction in feature representation, thereby affecting the performance of traditional segmentation methods. To address this issue, To address this issue, recent advancements in segmentation methods for small medical objects meticulously extract feature information from these objects using attention mechanisms. However, these methods mainly focus on general small medical objects and perform poorly in the segmentation of small polyps.Therefore, this study fully exploits the advantages of joint learning and multi-task learning. By guiding the high-resolution features to guide the low-resolution encoder through joint learning and fusing effective features from the super-resolution task to assist the segmentation task, our method enhances the model’s ability to extract texture details. We also design a high-resolution guidance module and an information complementation module, effectively integrating learning between different resolutions and tasks.Evaluation experiments on two settings on the Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB, CVC-300, and ETIS datasets demonstrate that our method achieves significant advantages in the segmentation of small polyps, outperforming various state-of-the-art methods.

Abstract Deformable multimodal image registration plays a key role in medical image analysis. It remains a challenge to find accurate dense correspondences between multimodal images due to the significant intensity distortion and the large deformation. macJNet is proposed to align the multimodal medical images, which is a weakly-supervised multimodal image deformable registration method using a joint learning framework and multi-sampling cascaded modality independent neighborhood descriptor (macMIND). The joint learning framework consists of a multimodal image registration network and two segmentation networks. The proposed macMIND is a modality-independent image structure descriptor to provide dense correspondence for registration, which incorporates multi-orientation and multi-scale sampling patterns to build self-similarity context. It greatly enhances the representation ability of cross-modal features in the registration network. The semi-supervised segmentation networks generate anatomical labels to provide semantics correspondence for registration, and the registration network helps to improve the performance of multimodal image segmentation by providing the consistency of anatomical labels. 3D CT-MR liver image dataset with 118 samples is built for evaluation, and comprehensive experiments have been conducted to demonstrate that macJNet achieves superior performance over state-of-the-art multi-modality medical image registration methods.

Under the multiple driving of capital, new media and consumer group, the output value of Chinese animation industry keeps growing rapidly. Meanwhile, the quality of animation characters image determines the overall development level of the animation industry. This paper proposed a detail feature preservation-based 3D facial expression transfer method aiming at the design of 3D characters. By extracting the detail features of 3D facial model, the basic expression after removing the details was obtained. By using the improved joint learning method, the basic expression of the source model was transferred to the target model, and the feature modulus based on the Laplace factor was constructed. Finally, through the detail feature vector adjustment strategy, the target model with source basic expression was performed detail restoration. The multiple experiments show that the method proposed in this paper can transfer the expression of the source model to the target model without damage, and at the same time, preserve the personality detail features of the target model. Moreover, the animation expressions generated are real and natural, and the facial micro-movements of animation characters are rich, which positively affects the development of Chinese animation industry.

The objective of event extraction is to recognize event triggers and event categories within unstructured text and produce structured event arguments. However, there is a common phenomenon of triggers and arguments of different event types in a sentence that may be the same word elements, which poses new challenges to this task. In this article, a joint learning framework for overlapping event extraction (ROPEE) is proposed. In this framework, a role pre-judgment module is devised prior to argument extraction. It conducts role pre-judgment by leveraging the correlation between event types and roles, as well as trigger embeddings. Experiments on the FewFC show that the proposed model outperforms other baseline models in terms of Trigger Classification, Argument Identification, and Argument Classification by 0.4%, 0.9%, and 0.6%. In scenarios of trigger overlap and argument overlap, the proposed model outperforms the baseline models in terms of Argument Identification and Argument Classification by 0.9%, 1.2%, 0.7%, and 0.6%, respectively, indicating the effectiveness of ROPEE in solving overlapping events. [ABSTRACT FROM AUTHOR]

Keypoint detection and description are the basis of many computer vision applications such as object recognition and image analysis. Current deep learning-based methods have made great progress in joint learning of keypoint detection and description construction. Low-level features have been proved to be helpful for keypoint detection and description. However, current detector and descriptor focus more on high-level feature and ignore the importance of low-level feature. They simply concatenate features and are lack of sufficient feature fusion. In this work, we propose a fusion representation learning network, which fuses different levels of features for both detectors and descriptors. Furthermore, we design and propose an adaptive feature fusion structure for the descriptor. Extensive experiments on HPatches, FM-Bench and Day-Night datasets demonstrate the superiority of our approach. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Southeast University / Dongnan Daxue Xuebao is the property of Journal of Southeast University 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.)

Multi-label ship recognition is a more challenging task than single-label ship recognition, mainly due to the complexity of the ship target location space. Environmental factors, such as inclement weather and uneven illumination, can negatively affect the acquisition of detailed features of each ship target. Unfortunately, existing multi-label ship recognition methods tend to focus more on high-level semantic features, neglecting the importance of low-level contour texture details in the target area. These methods also overlook the correlation between high-level semantic features and low-level contour texture details, resulting in insufficient information in the feature embedding space of the target area. To tackle these problems, we propose joint learning networks of low-level and high-level features for multi-label ship recognition in complex backgrounds(LHJL). Our proposed method consists of a circular attention guided(CAG) CNN module, which enriches the high-level semantic features of target regions and reduces the impact of complex environmental information. We also use low-level features after high-level deconvolution(AHDF) to output important contour texture details of the target area by weighting the key pixel values of the local detail features. Finally, joint learning of high-level features outputted by circular attention guides the CNN, and low-level features yielded by covariance pooling reduce the impact of complex environments, which can improve the mapping ability of detailed features of each ship target in complex spatial positions and complete multi-label ship recognition in complex backgrounds. We conducted experiments on public datasets such as Seaships and SMD and collected images of complex natural environments containing 59 ship categories. The results show that our proposed LHJL method outperforms other existing methods. [ABSTRACT FROM AUTHOR]

Point-level temporal action localization (PTAL) aims to locate action instances in untrimmed videos with only one timestamp annotation for each action instance. Existing methods adopt the localization-by-classification paradigm to locate action boundaries in the temporal class activation map (TCAM) by thresholding, also known as TCAM-based method. However, TCAM-based methods are limited by the gap between classification and localization tasks, since TCAM is generated by a classification network. To address this issue, we propose a re-training framework for the PTAL task, also known as LPR. This framework consists of two stages: pseudo-label generation and re-training. In the pseudo-label generation stage, we propose a feature embedding module based on a transformer encoder to capture global context features and optimize pseudo-labels' quality by leveraging point-level annotations. In the re-training stage, LPR uses the above pseudo-labels as supervision to locate action instances with a temporal action localization network rather than generating TCAMs. Furthermore, to alleviate the effects of label noise in the pseudo-labels, we propose a joint learning classification module (JLCM) in the re-training stage. This module contains two classification sub-modules that simultaneously predict action categories and are guided by a jointly determined clean set for network training. The proposed framework achieves state-of-the-art localization performance on both the THUMOS'14 and BEOID datasets. [ABSTRACT FROM AUTHOR]

Copyright of Journal of National University of Defense Technology / Guofang Keji Daxue Xuebao is the property of NUDT Press 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.)

Depression is a critical problem in modern society that affects an estimated 350 million people worldwide, causing feelings of sadness and a lack of interest and pleasure. Emotional disorders are gaining interest and are closely entwined with depression, because one contributes to an understanding of the other. Despite the achievements in the two separate tasks of emotion recognition and depression detection, there has not been much prior effort to build a unified model that can connect these two tasks with different modalities, including multimedia (text, audio, and video) and unobtrusive physiological signals (e.g., electroencephalography). We propose a novel temporal convolutional transformer with knowledge embedding to address the joint task of depression detection and emotion recognition. This approach not only learns multimodal embeddings across domains via the temporal convolutional transformer but also exploits special-domain knowledge from medical knowledge graphs to improve the performance of detection and recognition. It is essential that the features learned by our method can be perceived as a priori and are suitable for increasing the performance of other related tasks. Our method illustrates the case of “two birds with one stone” in the sense that two or more tasks can be efficiently handled with our unique model, which captures effective features. Experimental results on ten real-world datasets show that the proposed approach significantly outperforms other state-of-the-art approaches. On the other hand, experiments in which our methodology is applied to other reasoning tasks show that our approach effectively supports model reasoning related to emotion and improves its performance. [ABSTRACT FROM AUTHOR]

Deformable multimodal image registration plays a key role in medical image analysis. It remains a challenge to find accurate dense correspondences between multimodal images due to the significant intensity distortion and the large deformation. macJNet is proposed to align the multimodal medical images, which is a weakly-supervised multimodal image deformable registration method using a joint learning framework and multi-sampling cascaded modality independent neighborhood descriptor (macMIND). The joint learning framework consists of a multimodal image registration network and two segmentation networks. The proposed macMIND is a modality-independent image structure descriptor to provide dense correspondence for registration, which incorporates multi-orientation and multi-scale sampling patterns to build self-similarity context. It greatly enhances the representation ability of cross-modal features in the registration network. The semi-supervised segmentation networks generate anatomical labels to provide semantics correspondence for registration, and the registration network helps to improve the performance of multimodal image segmentation by providing the consistency of anatomical labels. 3D CT-MR liver image dataset with 118 samples is built for evaluation, and comprehensive experiments have been conducted to demonstrate that macJNet achieves superior performance over state-of-the-art multi-modality medical image registration methods. [ABSTRACT FROM AUTHOR]

Underwater object detection (UOD) has attracted widespread attention, being of great significance for marine resource management, underwater security and defense, underwater infrastructure inspection, etc. However, high-quality UOD tasks often encounter challenges such as image quality degradation, complex backgrounds, and occlusions between objects at different scales. This paper presents a collaborative framework for UOD via joint image enhancement and super-resolution to address the above problems. Specifically, a joint-oriented framework is constructed incorporating underwater image enhancement and super-resolution techniques. The proposed framework is capable of generating a detection-favoring appearance to provide more visual cues for UOD tasks. Furthermore, a plug-and-play self-attention mechanism, termed multihead blurpooling fusion network (MBFNet), is developed to capture sufficient contextual information by focusing on the dependencies between multiscale feature maps, so that the UOD performance of our proposed framework can be further facilitated. A comparative study on the popular URPC2020 and Brackish datasets demonstrates the superior performance of our proposed collaborative framework, and the ablation study also validates the effectiveness of each component within the framework. [ABSTRACT FROM AUTHOR]

Abstract Background Immigrant women’s challenges in realizing sexual and reproductive health and rights (SRHR) are exacerbated by the lack of knowledge regarding how to tailor post-partum contraceptive services to their needs. Therefore, the overall aim of the IMPROVE-it project is to promote equity in SRHR through improvement of contraceptive services with and for immigrant women, and, thus, to strengthen women’s possibility to choose and initiate effective contraceptive methods post-partum. Methods This Quality Improvement Collaborative (QIC) on contraceptive services and use will combine a cluster randomized controlled trial (cRCT) with a process evaluation. The cRCT will be conducted at 28 maternal health clinics (MHCs) in Sweden, that are the clusters and unit of randomization, and include women attending regular post-partum visits within 16 weeks post birth. Utilizing the Breakthrough Series Collaborative model, the study’s intervention strategies include learning sessions, action periods, and workshops informed by joint learning, co-design, and evidence-based practices. The primary outcome, women’s choice of an effective contraceptive method within 16 weeks after giving birth, will be measured using the Swedish Pregnancy Register (SPR). Secondary outcomes regarding women’s experiences of contraceptive counselling, use and satisfaction of chosen contraceptive method will be evaluated using questionnaires completed by participating women at enrolment, 6 and 12 months post enrolment. The outcomes including readiness, motivation, competence and confidence will be measured through project documentation and questionnaires. The project’s primary outcome involving women’s choice of contraceptive method will be estimated by using a logistic regression analysis. A multivariate analysis will be performed to control for age, sociodemographic characteristics, and reproductive history. The process evaluation will be conducted using recordings from learning sessions, questionnaires aimed at participating midwives, intervention checklists and project documents. Discussion The intervention’s co-design activities will meaningfully include immigrants in implementation research and allow midwives to have a direct, immediate impact on improving patient care. This study will also provide evidence as to what extent, how and why the QIC was effective in post-partum contraceptive services. Trial registration NCT05521646, August 30, 2022.

ABSTRACTAspect-based sentiment analysis (ABSA) has attracted many researchers' attention in recent years. However, the lack of benchmark datasets for specific languages is a common challenge because of the prohibitive cost of manual annotation. The zero-shot cross-lingual strategy can be applied to solve this gap in research. Moreover, previous works mainly focus on improving the performance of supervised ABSA with pre-trained languages. Therefore, there are few to no systematic comparisons of the benefits of multilingual models in zero-shot and joint training cross-lingual for the ABSA task. In this paper, we focus on the zero-shot and joint training cross-lingual transfer task for the ABSA. We fine-tune the latest pre-trained multilingual language models on the source language, and then it is directly predicted in the target language. For the joint learning scenario, the models are trained on the combination of multiple source languages. Our experimental results show that (1) fine-tuning multilingual models achieve promising performances in the zero-shot cross-lingual scenario; (2) fine-tuning models on the combination training data of multiple source languages outperforms monolingual data in the joint training scenario. Furthermore, the experimental results indicated that choosing other languages instead of English as the source language can give promising results in the low-resource languages scenario.

Abstract Emotion cause extraction (ECE) task that aims at extracting potential trigger events of certain emotions has attracted extensive attention recently. However, current work neglects the implicit emotion expressed without any explicit emotional keywords, which appears more frequently in application scenarios. The lack of explicit emotion information makes it extremely hard to extract emotion causes only with the local context. Moreover, an entire event is usually across multiple clauses, while existing work merely extracts cause events at clause level and cannot effectively capture complete cause event information. To address these issues, the events are first redefined at the tuple level and a span‐based tuple‐level algorithm is proposed to extract events from different clauses. Based on it, a corpus for implicit emotion cause extraction that tries to extract causes of implicit emotions is constructed. The authors propose a knowledge‐enriched joint‐learning model of implicit emotion recognition and implicit emotion cause extraction tasks (KJ‐IECE), which leverages commonsense knowledge from ConceptNet and NRC_VAD to better capture connections between emotion and corresponding cause events. Experiments on both implicit and explicit emotion cause extraction datasets demonstrate the effectiveness of the proposed model.

Existing vehicle re-identification (Re-ID) methods either extract valuable background information to enhance the robustness of the vehicle model or segment background interference information to learn vehicle fine-grained information. However, these methods do not consider the background information as a trade-off attribute to unite valuable background and background interference. This work proposes the trade-off background joint learning method for unsupervised vehicle Re-ID, which consists of two branches, to exploit the ambivalence of background information. In the global branch, a background focus of the pyramid global branch module is proposed to optimize the sample feature space. The designed pyramid background-aware attention extracts background-related features from the global image and constructs a two-fold confidence metric based on background-related and identity-related confidence scores to obtain robust clustering results during the clustering. In the local branch, a background filtering of the local branch module is proposed to alleviate the background interference. First, the background of each local region is segmented and weakened. Then, a background adaptive local label smoothing is designed to reduce noise in every local region. Comprehensive experiments on VeRi-776 and VeRi-Wild are conducted to validate the performance of the proposed balanced background information method. Experimental results show that the proposed method outperforms the state-of-the-art. [ABSTRACT FROM AUTHOR]