Task Characteristics Research Articles

Mitigating the Impact of Lognormal Atmospheric Turbulence Channel Estimation on FSO Communication Systems Using Advanced Deep Learning Modules

ABSTRACTCurrently, researchers and commercial entities worldwide are highly interested in optical wireless communication links operating in terrestrial environments. These links offer the distinct advantage of enabling high‐speed data transmission while keeping operational and installation costs low, all without requiring licensing. Free‐space optical (FSO) communication has established itself as a reliable method for delivering ultra‐high data‐rate services. However, the presence of atmospheric turbulence‐induced fading poses a significant challenge for FSO communication systems, particularly when operating over channels affected by time‐varying turbulence. The effectiveness of FSO systems is greatly influenced by atmospheric conditions in a specific area because the laser beam propagates through the atmosphere. As a consequence, this can lead to a significant degradation such as high complexity, high estimation error, and high bit error rate (BER) in the performance of FSO systems. This paper presents a stacked dual attention LSTM network with layered neural Turing machine (SDALSTM‐LaNTM) for FSO channel estimation over lognormal turbulence‐induced fading channels. In this approach, the stacked structure ensures a robust feature extraction capacity of SDALSTM and enhances its ability to capture intricate relationships among multiple atmospheric variables. After employing the stacked LSTM networks, the extracted features are combined with the augmented features obtained through the dual attention (DAttn) mechanism by concatenation. Then the LaNTM architecture used external memory to solve the channel parameter estimation problem and update new set of hidden features for estimation task. The ability to access and update memory enables the model to capture long‐term dependencies and context, making it beneficial for channel prediction in FSO systems. Presently, we are in the process of verifying the channel estimation model using measured FSO channel data. Simulation results clearly demonstrate the outstanding estimation performance of the SDALSTM‐LaNTM model in FSO systems, specifically when dealing with lognormal turbulence‐induced fading channels. Moreover, we highlight the effectiveness of the SDALSTM‐LaNTM model by conducting a comprehensive comparison with existing models, considering factors such as average capacity performance, outage probability estimation, and BER.

MSSA: Multi-Representation Semantics-Augmented Set Abstraction for 3D Object Detection

Accurate recognition and localization of 3D objects is a fundamental research problem in 3D computer vision. Benefiting from transformation-free point cloud processing and flexible receptive fields, point-based methods have become accurate in 3D point cloud modeling, but still fall behind voxel-based competitors in 3D detection. We observe that the set abstraction module, commonly utilized by point-based methods for downsampling points, tends to retain excessive irrelevant background information, thus hindering the effective learning of features for object detection tasks. To address this issue, we propose MSSA, a Multi-representation Semantics-augmented Set Abstraction for 3D object detection. Specifically, we first design a backbone network to encode different representation features of point clouds, which extracts point-wise features through PointNet to preserve fine-grained geometric structure features, and adopts VoxelNet to extract voxel features and BEV features to enhance the semantic features of key points. Second, to efficiently fuse different representation features of keypoints, we propose a Point feature-guided Voxel feature and BEV feature fusion (PVB-Fusion) module to adaptively fuse multi-representation features and remove noise. At last, a novel Multi-representation Semantic-guided Farthest Point Sampling (MS-FPS) algorithm is designed to help set abstraction modules progressively downsample point clouds, thereby improving instance recall and detection performance with more important foreground points. We evaluate MSSA on the widely used KITTI dataset and the more challenging nuScenes dataset. Experimental results show that compared to PointRCNN, our method improves the AP of “moderate” level for three classes of objects by 7.02%, 6.76%, and 5.44%, respectively. Compared to the advanced point-voxel-based method PV-RCNN, our method improves the AP of “moderate” level by 1.23%, 2.84%, and 0.55% for the three classes, respectively.

ViCo: Human visual perception-based contour extraction of Chinese Tang dynasty textile patterns

The study of ancient textile patterns not only enriches historical material but also inspires modern costume art design. Recent developments in computer vision technology provide novel analytical methods for the study of ancient textile patterns. As contours are one of the most significant features in computer vision tasks, contour extraction serves as a prerequisite to accomplishing intelligent understanding and automatic design of textile patterns. In this paper, a perception-consistent contour extraction method (ViCo) that simulates the human visual system from the retina, the lateral geniculate nucleus (LGN), to the V1 area in the cortex is proposed, along with an ancient textile pattern dataset represented by TuanKe in the Chinese Tang dynasty. Experiments were conducted both qualitatively and quantitatively to verify the effectiveness of the proposed method on the TuanKe dataset. Analysis of the results implies the effectiveness of ViCo because the retinal and LGN filter enhances edge information and restrains tiny pixel changes, and the cortex simulation is direction selective. This study is among the initial efforts to employ computer vision technology for ancient textile pattern studies. The study presents a TuanKe dataset and a novel bio-inspired contour extraction method, which has significant historical research importance, fabric art design inspiration, and computer application value.

Identification of Relationship between Meta Cognition and Misconceptions

This study investigates the complex interplay between meta cognition and misconceptions within the educational context, focusing on implications for effective teaching and learning strategies. Aligning with India's National Education Policy (NEP) 2020, which advocates for a transition from rote memorization to in- depth understanding, this research explores the concurrent presence of misconceptions and meta cognition in elementary science education. Meta cognition, the process of reflecting on one's cognitive processes, and misconceptions, or predefined notions that impede authentic learning, are analyzed through influential frameworks, including those developed by Flavell and Nelson and Narens. The MASRL model by Efklides is also incorporated to understand the interplay of meta cognitive knowledge, control strategies, experiences, task characteristics, and learning outcomes. The study highlights the impact of misconceptions as cognitive barriers and their influence on meta cognition, emphasizing the need for effective educational strategies. The literature review explores existing research on meta cognition and misconceptions, showcasing how meta cognition acts as a cognitive mirror, enabling learners to navigate and rectify misconceptions. This research has significant educational implications, highlighting the importance of cultivating self-directed learning, critical thinking, and evidence- based study strategies to enhance academic outcomes. The study concludes by emphasizing the dynamic interplay between meta cognition and misconceptions, offering insights for educators and researchers to design interventions fostering accurate conceptual understanding in learners.

Human factors considerations of Interaction between wearers and intelligent lower-limb prostheses: a prospective discussion.

Compared to traditional lower-limb prostheses (LLPs), intelligent LLPs are more versatile devices with emerging technologies, such as microcontrollers and user-controlled interfaces (UCIs). As emerging technologies allow a higher level of automation and more involvement from wearers in the LLP setting adjustments, the previous framework established to study human factors elements that affect wearer-LLP interaction may not be sufficient to understand the new elements (e.g., transparency) and dynamics in this interaction. In addition, the increased complexity of interaction amplifies the limitations of the traditional evaluation approaches of wearer-LLP interaction. Therefore, to ensure wearer acceptance and adoption, from a human factors perspective, we propose a new framework to introduce elements and usability requirements for the wearer-LLP interaction. This paper organizes human factors elements that appear with the development of intelligent LLP technologies into three aspects: wearer, device, and task by using a classic model of the human-machine systems. By adopting Nielsen's five usability requirements, we introduce learnability, efficiency, memorability, use error, and satisfaction into the evaluation of wearer-LLP interaction. We identify two types of wearer-LLP interaction. The first type, direct interaction, occurs when the wearer continuously interacts with the intelligent LLP (primarily when the LLP is in action); the second type, indirect interaction, occurs when the wearer initiates communication with the LLP usually through a UCI to address the current or foreseeable challenges. For each type of interaction, we highlight new elements, such as device transparency and prior knowledge of the wearer with the UCI. In addition, we redefine the usability goals of two types of wearer-LLP interaction with Nelson's five usability requirements and review methods to evaluate the interaction. Researchers and designers for intelligent LLPs should consider the new device elements that may additionally influence wearers' acceptance and the need to interpret findings within the constraints of the specific wearer and task characteristics. The proposed framework can also be used to organize literature and identify gaps for future directions. By adopting the holistic usability requirements, findings across empirical studies can be more comparable. At the end of this paper, we discuss research trends and future directions in the human factors design of intelligent LLPs.

Geometric Neural Operators (GNPs) for data-driven deep learning in non-euclidean settings

Abstract We introduce Geometric Neural Operators (GNPs) for data-driven deep learning of geometric features for tasks in non-euclidean settings. We present a formulation for accounting for geometric contributions along with practical neural network architectures and factorizations for training. We then demonstrate how GNPs can be used (i) to estimate geometric properties, such as the metric and curvatures of surfaces, (ii) to approximate solutions of geometric partial differential equations (PDEs) on manifolds, and (iii) to solve Bayesian inverse problems for identifying manifold shapes. These results show a few ways GNPs can be used for incorporating the roles of geometry in the data-driven learning of operators.

“Which Proof for Which Class?” Factors Influencing the Selection of Mathematical Proofs for Teaching

Selecting a proof for teaching is a frequent task for teachers. However, it is so far unclear, which factors are considered by teachers when selecting proofs. Is the selection based on task and proof characteristics such as the didactical type of proof? Or based on class characteristics such as students’ algebraic skills? Or do teachers’ characteristics such as their proof skills govern their decision? Or is the selection too non-generic for these characteristics to show a meaningful impact?To address these questions, a quasi-experimental study with N = 183 pre-service teachers was conducted to evaluate the influence of each of these factors on their selection of proofs for teaching. Results highlight several significant effects of the abovementioned characteristics and underline that—even at the pre-service level—the selection of proofs is more nuanced than often assumed in prior research and that teachers deliberately and adaptively select proofs for their teaching based on these factors.

Shared leadership strategies for languages education. A case of project-based learning at CLC in Poznan University of Technology

Abstract This article explains how shared leadership can be recognised and established in project-based learning as a situated English practice. It specifically explores the task characteristics of Cox et al.’s (2003) leadership model, which illustrates the functional relations between the procedural features of project work and its outcomes. The result of the analysis is that project work underlies the collective and highly integrative potential of the group of students, transforming them into a community of practice. The findings indicate that factors contributing to facilitating shared leadership are the interdependence of tasks between individuals driven by intrinsic motivation to act and the low complexity of tasks. The analysis also suggests that, although low complexity tasks may weaken shared leadership, they maximise relevance, thereby minimising cognitive effort. Consequently, this response strategy can help boost collaboration.

ACGAN for Addressing the Security Challenges in IoT-Based Healthcare System

The continuous evolution of the IoT paradigm has been extensively applied across various application domains, including air traffic control, education, healthcare, agriculture, transportation, smart home appliances, and others. Our primary focus revolves around exploring the applications of IoT, particularly within healthcare, where it assumes a pivotal role in facilitating secure and real-time remote patient-monitoring systems. This innovation aims to enhance the quality of service and ultimately improve people’s lives. A key component in this ecosystem is the Healthcare Monitoring System (HMS), a technology-based framework designed to continuously monitor and manage patient and healthcare provider data in real time. This system integrates various components, such as software, medical devices, and processes, aimed at improvi1g patient care and supporting healthcare providers in making well-informed decisions. This fosters proactive healthcare management and enables timely interventions when needed. However, data transmission in these systems poses significant security threats during the transfer process, as malicious actors may attempt to breach security protocols.This jeopardizes the integrity of the Internet of Medical Things (IoMT) and ultimately endangers patient safety. Two feature sets—biometric and network flow metric—have been incorporated to enhance detection in healthcare systems. Another major challenge lies in the scarcity of publicly available balanced datasets for analyzing diverse IoMT attack patterns. To address this, the Auxiliary Classifier Generative Adversarial Network (ACGAN) was employed to generate synthetic samples that resemble minority class samples. ACGAN operates with two objectives: the discriminator differentiates between real and synthetic samples while also predicting the correct class labels. This dual functionality ensures that the discriminator learns detailed features for both tasks. Meanwhile, the generator produces high-quality samples that are classified as real by the discriminator and correctly labeled by the auxiliary classifier. The performance of this approach, evaluated using the IoMT dataset, consistently outperforms the existing baseline model across key metrics, including accuracy, precision, recall, F1-score, area under curve (AUC), and confusion matrix results.

Multi-UAVs task allocation method based on MPSO-SA-DQN

Multi-UAVs play an important role in the battlefield. Although many methods are proposed to solve the Multi-UAV task allocation, there still existing the problems of complex time constraints and uncertain solution space. The reason is that multi-UAVs usually face changing environmental factors. Aiming at solving such problem, this paper proposes a multi-UAV task assignment method based on Deep Q-based evolutionary reinforcement learning algorithms (MPSO-SA-DQN). Specifically, this method builds a multi-agent training framework based on the deep evolutionary reinforcement learning mechanism and SA-DQN. Its aim is to improve the global exploration and optimization capabilities of multi-agents. At the same time, the multi-dimensional particle swarm optimization algorithm is introduced to optimize the state space. Based on task priority mapping, the MPSO-SA-DQN algorithm framework is proposed. As a result, multi-agents can optimize the execution state in real time in the environment interaction. Besides, it also has the ability to reach optimal state and maximum reward. According to the characteristics of multi-UAV global task assignment, this paper designs a priority state space autoencoder strategy and global task feature. A multi-UAVs tasks allocation and iterative optimization method based on MPSO-SA-DQN algorithm is proposed, so as to continuously optimize the task allocation scheme. The simulation results show that the multi-UAV task allocation method based on MPSO-SA-DQN can effectively solve the problem of uncertainty in the optimal solution space of task allocation. At the same time, the algorithm achieves faster convergence result, and a good prospect of promotion in the field of UAV swarm cooperative task planning.

Estimation of Speech Features Using a Wearable Inertial Sensor

Speech features have been investigated as novel digital biomarkers for many psychiatric and neurocognitive diseases. Microphones are the most used devices for speech recording but inevitably suffering from several disadvantages such as privacy leakage and environmental noises, limiting their clinical applications particularly for long-term ambulatory monitoring. The aim of the present study is therefore to explore the feasibility of extracting speech features from the acceleration recorded on the sternum. Ten healthy subjects volunteered in our study. Two speech tasks, i.e., repeating one sentence 20 times and reading 20 different sentences, were performed by each subject, with each task repeated 8 times under different speech rate and loudness. Voice signals and speech-caused chest vibrations were simultaneously recorded by a microphone and an accelerometer placed on the sternum. 42 acoustic features and 6 time-related prosodic features were extracted from both signals using a standard toolbox, and then compared by a linear fit and correlation analysis. Good agreement between the acceleration features and microphone features is observed in all 6 time-related prosodic features for both tasks, but only in 19 and 17 acoustic features for task 1 and 2, respectively, with most of them loudness- or pitch-related. Our results suggest the sternum acceleration to track time-related speech prosody, loudness, and pitch very well, demonstrating the feasibility of deriving digital biomarkers from the acceleration signal for diseases strongly related to time-related prosodic and loudness features.

Lock Operations Through the Operator’s Eyes: A Qualitative Exploration of Gaze Strategies

This paper reports a qualitative exploration of gaze strategies during closed-circuit television (CCTV) tasks in remote nautical object control of a lock. Previous research has not examined gaze strategies in scenarios where systems, such as nautical objects, are operated remotely using CCTV. As contextual factors matter in nautical object control, a qualitative approach was necessary to uncover domain-specific terminology and insights into gaze strategies. We recorded eye gaze from professional lock operators and then conducted semi-structured interviews with domain experts to assess these recordings. Thematic analysis revealed that experts were able to identify (features of) gaze strategies but did not share the same terminology. Based on this analysis we defined four strategies: anticipating, verifying, overview, and movement-directed gazes (RQ1). All strategies, except movement-directed gaze, were observed consistently across operators (RQ2), with verifying gaze aligning with task steps in a predefined protocol (RQ3). More generally, our classification framework from thematic analysis could help to systematically define and verify gaze strategies based on domain and task features across various CCTV working contexts. For nautical object control, the framework can be instrumental in interpreting and verifying future (quantitative) eye tracking results and informing instructional procedures.

A multi-task graph deep learning model to predict drugs combination of synergy and sensitivity scores

BackgroundDrug combination treatments have proven to be a realistic technique for treating challenging diseases such as cancer by enhancing efficacy and mitigating side effects. To achieve the therapeutic goals of these combinations, it is essential to employ multi-targeted drug combinations, which maximize effectiveness and synergistic effects.ResultsThis paper proposes ‘MultiComb’, a multi-task deep learning (MTDL) model designed to simultaneously predict the synergy and sensitivity of drug combinations. The model utilizes a graph convolution network to represent the Simplified Molecular-Input Line-Entry (SMILES) of two drugs, generating their respective features. Also, three fully connected subnetworks extract features of the cancer cell line. These drug and cell line features are then concatenated and processed through an attention mechanism, which outputs two optimized feature representations for the target tasks. The cross-stitch model learns the relationship between these tasks. At last, each learned task feature is fed into fully connected subnetworks to predict the synergy and sensitivity scores.The proposed model is validated using the O’Neil benchmark dataset, which includes 38 unique drugs combined to form 17,901 drug combination pairs and tested across 37 unique cancer cells. The model’s performance is tested using some metrics like mean square error (MSE\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$MSE$$\\end{document}), mean absolute error (MAE\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$MAE$$\\end{document}), coefficient of determination (R2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$${R}^{2}$$\\end{document}), Spearman, and Pearson scores. The mean synergy scores of the proposed model are 232.37, 9.59, 0.57, 0.76, and 0.73 for the previous metrics, respectively. Also, the values for mean sensitivity scores are 15.59, 2.74, 0.90, 0.95, and 0.95, respectively.ConclusionThis paper proposes an MTDL model to predict synergy and sensitivity scores for drug combinations targeting specific cancer cell lines. The MTDL model demonstrates superior performance compared to existing approaches, providing better results.

Adaptive Online Continual Learning for In-Situ Quality Prediction in Manufacturing Processes

Abstract Manufacturing processes undergo continuous changes in order to meet various requirements, such as process/product changes and variations in tool/workpiece conditions, leading to mixed, heterogenous, or anomalous data. As a result, a quality prediction model trained from previous data may not perform well when new tasks emerge. In order to achieve in-time and accurate product quality prediction, it is crucial to develop a predictive method that adapt to variations in the manufacturing system, capable of learning from new tasks without forgetting previous ones and detecting unknown tasks. This study proposes a deep learning method integrated with continual learning for in-situ quality prediction that is capable of learning from new tasks without forgetting previous ones. To demonstrate this idea, deep CNNs are designed to analyze in-process sensor data, which consist of shared layers to capture the common underlying features across all tasks, and task-specific layers that capture specific characteristics of each individual task. In order to identify the task to which incoming product belongs, a task prediction approach based on task relevancy using filter subspace distance is proposed. When new data come in, the model first identifies the task, followed by predicting the quality of the current product. The proposed method is demonstrated on two case studies, including quality prediction of the workpiece using acoustic emissions during the laser-induced plasma micro-machining process, and quality prediction of the product through thermal images during the hot stamping process.

The Use of Language Sample Analysis to Differentiate Developmental Language Disorder From Typical Language in Bilingual Children: A Systematic Review and Meta-Analysis.

Language sample analysis (LSA) is a commonly recommended method of assessment for bilingual children. This systematic review and meta-analysis provides a comprehensive overview of the literature on the use of LSA to differentiate between developmental language disorder (DLD) and typical language (TL) in bilingual children. We conducted a search of several large electronic databases along with forward and backward searches and applied abstract and full-text screening procedures to identify all relevant studies. We then estimated standardized mean differences, representing the ability of LSA to differentiate between DLD and TL, using multilevel model and subgroup and moderator analyses to identify characteristics of LSA that may be associated with differences in effect size magnitude. We conducted assessments of publication bias and risk of bias by examining quality indicators for each study. The search yielded 35 articles that met the inclusion criteria. Participants ranged in age from 2;0 (years;months) to 11;9, with over 40 languages represented. Across studies, the pooled standardized mean difference indicated that children with DLD performed 0.78 SD lower on LSA measures than those with TL. Measures of morphosyntactic accuracy exhibited the largest pooled effect size. Elicitation method, language of task, and age were not associated with differences in effect size. Results of this study provide evidence of the clinical utility of LSA in differentiating between DLD and TL in bilingual children. Further research is needed to examine classification accuracy as well as task characteristics that may improve its diagnostic utility.

Wind Turbine Bearing Failure Diagnosis Using Multi-Scale Feature Extraction and Residual Neural Networks with Block Attention

Wind turbine rolling bearings are crucial components for ensuring the reliability and stability of wind power systems. Their failure can lead to significant economic losses and equipment downtime. Therefore, the accurate diagnosis of bearing faults is of great importance. Although existing deep learning fault diagnosis methods have achieved certain results, they still face limitations such as inadequate feature extraction capabilities, insufficient generalization to complex working conditions, and ineffective multi-scale feature capture. To address these issues, this paper proposes an advanced fault diagnosis method named the two-stream feature fusion convolutional neural network (TSFFResNet-Net). Firstly, the proposed method combines the advantages of one-dimensional convolutional neural networks (1D-ResNet) and two-dimensional convolutional neural networks (2D-ResNet). It transforms one-dimensional vibration signals into two-dimensional images through the empirical wavelet transform (EWT) method. Then, parallel convolutional kernels in 1D-ResNet and 2D-ResNet are used to extract multi-scale features, respectively. Next, the Convolutional Block Attention Module (CBAM) is introduced to enhance the network’s ability to capture key features by focusing on important features in specific channels or spatial areas. After feature fusion, CBAM is introduced again to further enhance the effect of feature fusion, ensuring that the features extracted by different network branches can be effectively integrated, ultimately providing more accurate input features for the classification task of the fully connected layer. The experimental results demonstrate that the proposed method outperforms other traditional methods and advanced convolutional neural network models on different datasets. Compared with convolutional neural network models such as LeNet-5, AlexNet, and ResNet, the proposed method achieves a significantly higher accuracy on the test set, with a stable accuracy of over 99%. Compared with other models, it shows better generalization and stability, effectively improving the overall performance of rolling bearing vibration signal fault diagnosis. The method provides an effective solution for the intelligent fault diagnosis of wind turbine rolling bearings.

Task Characteristics Associated with Mathematical Word Problem-Solving Performance Among Elementary School-Aged Children: A Systematic Review and Meta-Analysis

Mathematical word problem-solving skills are crucial for students across their lives, yet solving such tasks poses challenges for many. Therefore, understanding the characteristics of mathematical word problems that are associated with students’ performance is important. The objective of this systematic review and meta-analysis was to evaluate the effects of linguistic and numerical task characteristics associated with mathematical word problem-solving performance among elementary school-aged children (Grades 1 to 6). The systematic review was based on five electronic databases and citation searching. Reporting was conducted following The Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA). The findings (K = 69) showed that five of the six investigated linguistic task characteristics (i.e., the position of the unknown, schematic structure, irrelevant information, realistic considerations, and lexical consistency) and one of the two numerical task characteristics (i.e., number of operations) were related (g = 0.39 to 4.26) with elementary school-aged children’s mathematical word problem-solving. However, the findings did not provide support for a general association between a familiar situational narrative or the required operation with mathematical word problem-solving. The findings highlight that elementary school-aged children especially struggle with mathematical word problems requiring realistic considerations or multiple mathematical operations, containing lexical inconsistency, and problems in which the position of the unknown is the first value. This further understanding of elementary schoolers’ word problem-solving performance may guide the design of appropriate and progressive instruction and assessment tools and steer research into the interactions within task characteristics and with individual characteristics.

ETFE: Efficient transformer with feature enhancement for semantic segmentation

Abstract Despite the powerful capabilities of Transformers in relation to modeling, their effectiveness in capturing multi-scale object features in image-based tasks is often limited. This limitation arises from the diverse shapes and distances between objects in natural scenes, which pose challenges for Transformers to accurately capture object sizes and distances. To overcome this problem, we propose an effective Transformer which extracts multi-scale features at different resolutions. First, we leverage a multi-head attention mechanism to extract richer semantic features at various resolutions. The semantic features are then passed through the Feature Gate to aggregate multi-scale semantic features. Finally, Attentional Feature Enhancement further strengthens the semantic information of target features, enabling the modeling of dependencies among multi-scale features. Comprehensive experiments conducted on the Cityscapes and ADE20K datasets have shown that our model outperforms competing methods in both performance and accuracy.

Misalignment-resistant domain adaptive learning for one-stage object detection

Without consideration of task specificity, directly transforming domain adaptive pipelines from classification to one-stage detection tends to pose severer misalignments. These misalignments include: (1) Foreground misalignment that the domain discriminator obsessively concentrates on backgrounds since one-stage detectors do not contain proposals for instance-level discrimination. (2) Localization misalignment that domain-adaptive features supervised by the domain discriminator are not suitable for localization tasks, as the discriminator is a classifier in essence. To tackle these problems, we propose the Misalignment-Resistant Domain Adaption (MRDA) for one-stage detectors. Specifically, to alleviate foreground misalignment, a mask-based domain discriminator is proposed to perform instance-level discrimination by assigning the pixel-level domain labels based on instance-level masks. As for localization misalignment, a localization discriminator is introduced to learn domain-adaptive features for localization tasks. It employs an additional box-regression branch with an IoU loss to perform adversarial mutual supervision with the feature extractor. Comprehensive experiments demonstrate that our method effectively mitigates the misalignments and achieves state-of-the-art detection across multiple datasets.

Exploiting Neighbor Effect: Conv-Agnostic GNN Framework for Graphs With Heterophily.

Due to the homophily assumption in graph convolution networks (GCNs), a common consensus in the graph node classification task is that graph neural networks (GNNs) perform well on homophilic graphs but may fail on heterophilic graphs with many interclass edges. However, the previous interclass edges' perspective and related homo-ratio metrics cannot well explain the GNNs' performance under some heterophilic datasets, which implies that not all the interclass edges are harmful to GNNs. In this work, we propose a new metric based on the von Neumann entropy to reexamine the heterophily problem of GNNs and investigate the feature aggregation of interclass edges from an entire neighbor identifiable perspective. Moreover, we propose a simple yet effective Conv-Agnostic GNN framework (CAGNNs) to enhance the performance of most GNNs on the heterophily datasets by learning the neighbor effect for each node. Specifically, we first decouple the feature of each node into the discriminative feature for downstream tasks and the aggregation feature for graph convolution (GC). Then, we propose a shared mixer module to adaptively evaluate the neighbor effect of each node to incorporate the neighbor information. The proposed framework can be regarded as a plug-in component and is compatible with most GNNs. The experimental results over nine well-known benchmark datasets indicate that our framework can significantly improve performance, especially for the heterophily graphs. The average performance gain is 9.81%, 25.81%, and 20.61% compared with graph isomorphism network (GIN), graph attention network (GAT), and GCN, respectively. Extensive ablation studies and robustness analysis further verify the effectiveness, robustness, and interpretability of our framework. Code is available at https://github.com/JC-202/CAGNN.