检索结果-内蒙古大学图书馆

Surge-NF: Neural Fields inspired peak storm surge surrogate modeling with multi-task learning and positional encoding

引用

COASTAL ENGINEERING 2024年 193卷

作者： Jiang, Wenjun Zhong, Xi Zhang, Jize Hong Kong Univ Sci & Technol Dept Civil & Environm Engn Hong Kong Peoples R China

Storm surges pose a significant threat to coastal communities, necessitating rapid and precise storm surge prediction methods for long-time risk assessment and emergency management. High-fidelity numerical models such as ADCIRC provide accurate storm surge simulations but are computationally expensive. Surrogate models have emerged as an alternative option to alleviate the computational burden by learning from available numerical datasets. However, existing surrogate models face challenges in capturing the highly non-stationary and non-linear patterns of storm surges, resulting in over-smoothed response surfaces. Moreover, the dry-wet status of nearshore nodes has not been informatively considered in the training process. This study proposes Surge-NF, a novel point-based surrogate model inspired by Neural Fields (NF) from computer graphics. Surge-NF introduces two key innovations. A positional encoding module is proposed to mitigate over-smoothing of high-frequency peak storm surge spatial dependencies. A multi-task learning framework is proposed to simultaneously learn and predict the dry-wet status and peak surge values, leveraging task dependencies to improve prediction accuracy and data efficiency. We evaluate Surge-NF on the NACCS database with comparison to state-of-the-art alternative surrogate models. Surge-NF consistently reduces RMSE/MAE by 50% and achieves 4-5 times computational cost gain over baselines, requiring only 50 training storms to produce accurate predictions. The complementary benefits of the positional encoding and multi-task learning modules are evident from the improved prediction capability with their combined use. Overall, Surge-NF represents a significant advancement in storm surge surrogate modeling, offering its novel and unique ability to capture high-frequency spatial variations and leverage task dependencies. It has the potential to greatly enhance storm surge risk assessment and emergency response management, enabling effective dec

关键词： Storm surge Surrogate model Neural fields Deep learning Multi-task learning positional encoding

来源：评论

学校读者我要写书评

暂无评论

Structural and positional ensembled encoding for Graph Transformer

引用

PATTERN RECOGNITION LETTERS 2024年 183卷 104-110页

作者： Yeom, Jeyoon Kim, Taero Chang, Rakwoo Song, Kyungwoo Yonsei Univ Dept Stat & Data Sci Dept Appl Stat Seoul South Korea Univ Seoul Dept Appl Chem Seoul South Korea Univ Seoul Dept Artificial Intelligence Seoul South Korea

In the Transformer architecture, positional encoding is a vital component because it provides the model with information about the structure and position of data. In Graph Transformer, there have been attempts to introduce different positional encodings and inject additional structural information. Therefore, in terms of integrating positional and structural information, we propose a Structural and positional Ensembled Graph Transformer (SPEGT). We developed SPEGT by noting the different properties of structural and positional encodings of graphs and the similarity of their computational processes. We have set a unified component that integrates the functionalities: (i) Random Walk positional encoding, (ii) Shortest Path Distance between each node, and (iii) Hierarchical Cluster encoding. We find a problem with a well-known positional encoding and experimentally verify that combining it with other encodings can solve their problem. In addition, SPEGT outperforms previous models on a variety of graph datasets. We also show that SPEGT using unified positional encoding, performs well on structurally indistinguishable graph data through error case analysis.

关键词： Graph neural network Graph Transformer positional encoding Graph clustering Attention

来源：评论

学校读者我要写书评

暂无评论

A super-pixel slicing enhanced positional encoding for vision transformers

A super-pixel slicing enhanced positional encoding for visio...

引用

第43届中国控制会议

作者： Junchao Lu Xue-song Tang Kuangrong Hao College of Information Science and Technology Donghua University

ISBN: (数字)9789887581581

ISBN: (纸本)9798350366907

Transformers have performed exceptionally well in natural language processing(NLP), prompting researchers to study their potential in computer vision. One such transformer is the Vision Transformer(ViT), which uses a pure transformer structure to classify images through a sequence of fixed-size patches. However, relying on the same patch embedding method for all images is considered an oversimplification of the process. In response to this limitation, we propose a super-pixel slicing scheme that dynamically generates a sequence of patch embeddings based on an image's features. This sequence is then utilized to create a progressive positional encoding which can bring together slices of the same object. Our method can seamlessly integrate into the existing Transformer framework to form an end-to-end Vision Transformer with Superpixel Slicing(SSVT).Our empirical results show that SSVT significantly improves the performance of transformer-based models in image classification tasks.

关键词： Image Classification Vision Transformer Super-pixel Slicing Patch Embedding positional encoding

来源：评论

学校读者我要写书评

暂无评论

A hierarchical transformer-based network for multivariate time series classification

引用

INFORMATION SYSTEMS 2025年 132卷

作者： Tang, Yingxia Wei, Yanxuan Li, Teng Zheng, Xiangwei Ji, Cun Shandong Normal Univ Sch Informat Sci & Engn Jinan 250358 Shandong Peoples R China Shandong Univ Informatizat Off Jinan 250100 Shandong Peoples R China

In recent years, Transformer has demonstrated considerable potential in multivariate time series classification due to its exceptional strength in capturing global dependencies. However, as a generalized approach, it still faces challenges in processing time series data, such as insufficient temporal sensitivity and inadequate ability to capture local features. In this paper, a hierarchical Transformer-based network (Hformer) is innovatively proposed to address these problems. Hformer handles time series progressively at various stages to aggregate multi-scale representations. At the start of each stage, Hformer segments the input sequence and extracts features independently on each temporal slice. Furthermore, to better accommodate multivariate time series data, a more efficient absolute position encoding as well as relative position encoding are employed by Hformer. Experimental results on 30 multivariate time series datasets of the UEA archive demonstrate that the proposed method outperforms most state-of-the-art methods.

关键词： Multivariate time series classification Self-attention positional encoding

来源：评论

学校读者我要写书评

暂无评论

Research on vehicle trajectory prediction algorithm based on improved transformer

引用

PROCEEDINGS OF THE INSTITUTION OF MECHANICAL ENGINEERS PART D-JOURNAL OF AUTOMOBILE ENGINEERING 2025年

作者： Jin, Meihan Wang, Pengyu Wang, Weihua Zhang, Zhiyang Jiang, Lingshan Jilin Univ Coll Automot Engn Changchun 130025 Jilin Peoples R China Faw Volkswagen Automot Co Ltd Tech Dev Dept Changchun Jilin Peoples R China

Trajectory prediction is an important part in autonomous driving technology. The current prediction algorithm model still has some problems in the interaction behavior of traffic participants, the utilization rate of highly refined map information, and the learning of potential traffic rules in the real world. In this paper, a vehicle trajectory prediction algorithm model based on neural network is built, which selects Argoverse 2 Motion Forecasting as the training data set of the model. Relative spatiotemporal position encoding is used to encode the input information of the trajectory prediction model. It can better understand the complex interaction behavior in traffic scenes. The K-Means++ algorithm is used for the trajectory clustering of the Waymo dataset, integrating their results as prior knowledge into the trajectory prediction model to enhance their ability of learning from the potential traffic rules in the real world. The performance of the model was tested according to the selected data set. The average displacement error, the final displacement error, and its minimum value are used as the evaluating indicator of output trajectory and perform quantitative analysis. Visualizing the prediction trajectory of the model to visually judge the error and feasibility of the model. The results prove that the prediction model has good prediction ability for some behaviors such as going straight, turning left, and turning right. It can also well predict the future state information of other traffic participants such as pedestrians and non-motor vehicles.

关键词： Trajectory prediction neural networks positional encoding attention mechanism prior knowledge

来源：评论

学校读者我要写书评

暂无评论

Attentional modulation of peripheral pointing hypometria in healthy participants: An insight into optic ataxia?

引用

NEUROPSYCHOLOGIA 2025年 208卷 109084页

作者： Jurkiewicz, Tristan Vialatte, Audrey Yeshurun, Yaffa Pisella, Laure Univ Claude Bernard Lyon 1 Ctr Rech Neurosci Lyon CNRS UMR5292 INSERM U1028 F-69500 Bron France Ctr Explorat Retine Kleber CERK 50 Cours Franklin Roosevelt F-69006 Lyon France Univ Haifa Sch Psychol Sci Haifa Israel

Damage to the superior parietal lobule and intraparietal sulcus (SPL-IPS) causes optic ataxia (OA), characterized by pathological gaze-centered hypometric pointing to targets in the affected peripheral visual field. The SPL-IPS is also involved in covert attention. Here, we investigated the possible link between attention and action. This study investigated the effect of attention on pointing performance in healthy participants and two OA patients. In invalid trials, targets appeared unpredictably across different visual fields and eccentricities. Valid trials involved cued targets at specific locations. The first experiment used a central cue with 75% validity, the second used a peripheral cue with 50% validity. The effect of attention on pointing variability (noise) or time was expected as a confirmation of cueing efficiency. Critically, if OA reflects an attentional deficit, then healthy participants, in the invalid condition (without attention), were expected to produce the gaze-centered hypometric pointing bias characteristic of OA. Results: revealed main effects of validity on pointing biases in all participants with central predictive cueing, but not with peripheral low predictive cueing. This suggests that the typical underestimation of visual eccentricity in OA (visual field effect) at least partially results from impaired endogenous attention orientation toward the affected visual field.

关键词： Superior parietal lobule Intraparietal sulcus Covert attention Reaching positional encoding Peripheral vision Spatial attention

来源：评论

学校读者我要写书评

暂无评论

Neural Genome Sequence encoding for Downy Mildew Pathogen Detection: A Spatial Pyramid and Dual-Attention Approach

引用

IEEE ACCESS 2025年 13卷 65478-65489页

作者： Rani, Geeta Misra, Ankit Kundu, Nidhi Dhaka, Vijaypal Singh Lather, Pankaj Manipal Univ Jaipur Dept IoT & IS Jaipur 303007 India Jio Pvt Ltd Bengaluru 560103 Karnataka India Sri Karan Narendra Agr Univ Dept Stat Math & Comp Sci Jobner 303329 India Manipal Univ Jaipur Dept Comp & Commun Engn Jaipur 303007 India Bhai Parmanand DSEU Shakarpur Campus New Delhi 110092 India

A worldwide crop loss ranging from approximately 6 to 60% in pearl millet has been reported due to downy mildew caused by Sclerospora graminicola. This disease is transmitted through soil or seeds carrying oospores. Field screening, greenhouse experiments, and laboratory techniques are effective in preventing downy mildew in pearl millet;however, these methods are expensive, time-consuming, and require expertise in plant pathology. To address these challenges, researchers have turned to artificial intelligence, employing deep learning techniques such as long-short term memory, and recurrent neural networks, for disease identification using genomic sequences of pathogens. However, existing methods have limitations in interpreting lengthy genomic sequences, preserving the relative positioning of nucleotides, and identifying the minimum sequence length necessary to differentiate one genomic sequence from similar ones. Additionally, these models typically involve numerous training parameters and require significant computation time. Furthermore, none of the existing models specifically focus on recognizing Sclerospora graminicola. In this manuscript, we propose a Neural Genome Sequence Encoder (NeuGSE) model for early prediction of downy mildew in pearl millet using genomic sequences of the fungus Sclerospora graminicola. We also demonstrate the impact of genomic sequence length, positional encoding, and convolutional block attention modules on the performance of the proposed model. Furthermore, we apply two classifiers to prevent halt in training the model due to local minima. Thus, the model achieves the highest precision of 98.04% and an F1 score of 98% on genomic sequence of length 10,000. Moreover, the average F1 score of the proposed model is 19.53% and 13.81% higher than long-short term memory and bidirectional long-short term memory models, respectively, on the same dataset. Additionally, the model can recognize Sclerospora graminicola even in a genomic sequence

关键词： Genomics Bioinformatics Accuracy encoding Computational modeling Attention mechanisms Crops Pathogens Training Feature extraction Agriculture disease deep learning genetic resources crops pearl millet positional encoding

来源：评论

学校读者我要写书评

暂无评论

Hybrid Offset Position encoding for Large-Scale Point Cloud Semantic Segmentation

引用

REMOTE SENSING 2025年第2期17卷 256-256页

作者： Xiao, Yu Wu, Hui Chen, Yisheng Chen, Chongcheng Dong, Ruihai Lin, Ding Fuzhou Univ Acad Digital China Fuzhou 350108 Peoples R China Univ Coll Dublin Sch Comp Sci Dublin D04 V1W8 Ireland

In recent years, large-scale point cloud semantic segmentation has been widely applied in various fields, such as remote sensing and autonomous driving. Most existing point cloud networks use local aggregation to abstract unordered point clouds layer by layer. Among these, position embedding serves as a crucial step. However, current methods of position embedding have limitations in modeling spatial relationships, especially in deeper encoders where richer spatial positional relationships are needed. To address these issues, this paper summarizes the advantages and disadvantages of mainstream position embedding methods and proposes a novel Hybrid Offset Position encoding (HOPE) module. This module comprises two branches that compute relative positional encoding (RPE) and offset positional encoding (OPE). RPE combines explicit encoding to enhance position features through attention, learning position bias implicitly, while OPE calculates absolute position offset encoding by considering differences with grouping embeddings. These two encodings are adaptively mixed in the final output. The experiment conducted on multiple datasets demonstrates that our module helps the deep encoders of the network capture more robust features, thereby improving model performance on various baseline models. For instance, PointNet++ and PointMetaBase enhanced with HOPE achieved mIoU gains of 2.1% and 1.3% on the large-scale indoor dataset S3DIS area-5, 2.5% and 1.1% on S3DIS 6-fold, and 1.5% and 0.6% on ScanNet, respectively. RandLA-Net with HOPE achieved a 1.4% improvement on the large-scale outdoor dataset Toronto3D, all with minimal additional computational cost. PointNet++ and PointMetaBase had approximately only a 0.1 M parameter increase. This module can serve as an alternative for position embedding, and is suitable for point-based networks requiring local aggregation.

关键词： positional encoding position embedding local aggregation attention mechanism large-scale point cloud semantic segmentation

来源：评论

学校读者我要写书评

暂无评论

CFTformer: End-to-End Cross-Frame Multi-Object Tracking with Transformer

引用

IEEE Access 2025年 13卷 80587-80600页

作者： Amirkhani, Abdollah Khoshnevis, Seyed Alireza Iran University of Science and Technology School of Automotive Engineering Tehran 16846-13114 Iran University of South Florida Department of Electrical Engineering Tampa FL United States

Multi-object tracking (MOT) has been at the center of numerous applications from autonomous vehicles (AVs) to surveillance and even retail analytics. Traditional MOT methods typically rely on motion-based and appearance-based similarity information to associate detections across frames. However, the new transformer attention-based approach to MOT has removed the need for complex post-processing steps, such as graph optimization, allowing for end-to-end query tracking across frames. While the new transformer-based approaches offer many advantages, in the majority of these models the temporal dimension of the sequence is only considered in either the iterative processing of the frames or the memory of the queries. The proposed cross-frame multi-object tracking transformer (CFTforrmer) aims to improve one of the challenging areas of tracking, the association across different frames in the temporal dimension. In the proposed approach, the temporal identities of the frames are included in the positional encoding of the patches. This approach allows the encoder-decoder to track the queries more efficiently across the frames. For this model, scalable deformable-attention layers were used to design the encoder and decoder to decrease the computational cost. CFTformer also employs the proposed attention-based trajectory refinement (ATR) scheme to improve the tracking performance in blurred frames. The three-dimensional positional encoding of the patches helps the proposed ATR module to better capture the trajectories of the queries and generate smoother predictions. Overall, the model was able to achieve 1.7% and 0.6% improvement in the identification F1 score (IDF1) metric on MOT17 and MOT20 datasets while having ~0-15% lower number of identity switches, compared to other transformer-based approaches. More accurate tracking and lower identity switches make this algorithm more suitable to be used in the field of autonomous driving. To access model’s performance in AV applica

关键词： CFTformer multi-object tracking (MOT) positional encoding trajectory refinement transformer network

来源：评论

学校读者我要写书评

暂无评论

A New Monitoring Method for the Injection Volume of Blast Furnace Clay Gun Based on Object Detection

引用

PROCESSES 2025年第4期13卷 1006-1006页

作者： Zhang, Xunkai Liang, Helan Guo, Hongwei Yan, Bingji Xu, Hao Soochow Univ Sch Comp Sci Suzhou 215000 Peoples R China Soochow Univ Shagang Iron & Steel Coll Suzhou 215000 Peoples R China

Monitoring the injection volume of clay guns is important in blast furnace ironmaking. Currently, such monitoring data are often recorded manually, which has limitations such as low reliability and high delay. To address these issues, we revisit the task from a computer vision perspective and propose an object detection method. First, we introduce an interpolation annotation technique to build a clay gun dataset. With it as a foundation, an improved object detection model called Faster R-CNN with multi-stage positional encoding and two-branch Self Challenge (PSCfrcn) is proposed. Our model leverages multi-stage positional encoding (PE) to focus more on the relative position of local features, and a self-challenge (SC) module to mitigate the interference caused by the harsh environments. We conduct extensive experiments on the clay gun dataset from actual industrial scenarios, and use various metrics to validate the performance. Experimental results demonstrate that our method can significantly enhance the model's discriminative power and generalization ability, which is a promising direction for this task.

关键词： blast furnace clay injection object detection positional encoding self-challenge

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：