检索结果-内蒙古大学图书馆

DALO-APR: LLM-based automatic program repair with data augmentation and loss function optimization

JOURNAL OF SUPERCOMPUTING 2025年第5期81卷

作者： Wang, Shaosheng Lu, Lu Qiu, Shaojian Tian, Qingyan Lin, Haishan South China Univ Technol Sch Comp Sci & Engn Guangzhou 510006 Guangdong Peoples R China Guangdong Artificial Intelligence & Digital Econ L Guangzhou 510320 Guangdong Peoples R China Pengcheng Lab Shenzhen 518055 Guangdong Peoples R China South China Agr Univ Coll Math & Informat Guangzhou 510642 Guangdong Peoples R China Guangdong Prov Key Lab Tunnel Engn Safety & Emerge Guangzhou 510440 Guangdong Peoples R China

Automatic program repair (APR) has made significant strides with the advent of large language models (LLMs) such as T5 and CodeT5. However, LLM-based APR models may rely on repetitive repair patterns due to limited training data diversity, resulting in suboptimal performance. Additionally, common loss functions, such as cross-entropy, may not fully prioritize repair locations or optimize the model's output probability distribution to favor more accurate repair candidates. To address these challenges, this paper proposes a method for LLM-Based APR with Data Augmentation and loss function Optimization (DALO-APR). The data augmentation strategy expands the variety of repair patterns by randomly deleting, inserting, swapping tokens, and injecting errors. The optimized loss function helps the model rank more accurate repair candidates higher. Experimental results on Java, JavaScript, Python, and C datasets demonstrate that DALO-APR improves both error localization and bug fixing. Compared to baseline models, DALO-APR shows improvements across multiple metrics, especially with a 105.65% increase in 100% accuracy.

关键词： Automatic program repair Large language models Data augmentation loss function

来源：评论

学校读者我要写书评

暂无评论

A robust image watermarking framework based on U²-net encoder and loss function weight assignment

引用

MULTIMEDIA SYSTEMS 2025年第1期31卷 1-17页

作者： Xu, Kangkang Han, Wen Fang, Yixiang Zhao, Yi Li, Jun Wang, Junxiang Jingdezhen Ceram Univ Sch Mech & Elect Engn Jingdezhen 333403 Jiangxi Peoples R China

In the field of watermarking, the process of guiding the watermark embedding is crucial, which can determine the imperceptibility and robustness of the watermarking algorithms. Traditional algorithms for guided watermark embedding are susceptible to feature extraction capabilities and their robustness cannot meet the demands of practical applications. Recently, some watermarking frameworks based on deep learning have been proposed, which employ an encoder to automatically extract features and guide the watermark embedding, and show better robustness in practical applications. However, some encoders also suffer from insufficient feature extraction capability due to their network structure, resulting in unsatisfactory visual quality. To this end, an encoder based on improved U-2-Net structure is proposed, which performs multi-scale and multi-dimensional feature extraction on images to generate high-resolution feature maps with edge information. These feature maps help guide the watermark to be embedded in the optimal pixel space, thus improving the visual quality. In addition, a weight assignment strategy that combines square growth and linear growth is proposed to balance visual quality and robustness in training. The experimental results demonstrate that our method outperforms existing end-to-end watermarking schemes in terms of imperceptibility while ensuring high robustness.

关键词： Watermarking U-2-Net Weight assignment loss function

来源：评论

学校读者我要写书评

暂无评论

A novel loss function for early event detection based on infinite mixture prototypes for facial expressions

引用

PATTERN RECOGNITION LETTERS 2025年 190卷 59-65页

作者： Yang, Zhi-Fang Chu, Heng Natl Taipei Univ Dept Comp Sci & Informat Engn Taipei 237 Taiwan

Early event detection aims to identify events at their early stages, enabling timely interventions and responses. In our prior work (Wang et al., 2022), a primal estimated sub-gradient solver replaced quadratic programming in the max-margin early event detector (MMED) (Hoai and De la Torre, 2014), enhancing computational efficiency. In another study (Yang and Chiu, 2024), infinite mixture prototypes (IMP) (Allen et al., 2019), integrated with prototypical networks, were applied to early event detection, combining few-shot learning and multimodal data. MMED's loss function was adapted into a re-scaling function to incorporate early event information, though further refinement was required. In this work, we introduce a novel loss function for facial expression early event detection based on IMP. This function increases penalties for late detection and incorporates scalars to better account for the influence of early events on model performance. Experimental results show that the proposed loss function significantly improves accuracy while maintaining consistent event detection length.

关键词： Early event detection Infinite mixture prototypes Few-shot learning loss function

来源：评论

学校读者我要写书评

暂无评论

A Seagull loss function With Application to Recommender Systems

引用

IEEE TRANSACTIONS ON COMPUTATIONAL SOCIAL SYSTEMS 2025年

作者： Xu, Yuan-Yuan Xiao, Hui Zhang, Heng-Ru Wang, Xi-Zhao Yu, Hong Min, Fan Southwest Petr Univ Sch Comp Sci & Software Engn Chengdu 610500 Peoples R China Shenzhen Univ Inst Big Data Shenzhen 518060 Peoples R China Chongqing Univ Posts & Telecommun Coll Comp Sci & Technol Chongqing 400065 Peoples R China

Outliers are intrinsic to recommender systems (RSs) due to user uncertainty and bring large deviations to the total loss. Classical loss functions such as L-1 and L-2 do not consider this issue, thus biasing the model toward abnormal ratings. Advanced ones such as Logcosh, Huber, and Wing losses try to control model biasing;however, they do not consider the rating ranges of recommender systems. In this article, we propose a Seagull loss combining quadratic and softsign functions to handle this issue. The quadratic function aims to preserve or appropriately increase losses caused by small deviations, while softsign function aims to suppress losses caused by large deviations. It is distinguished from three types of popular losses. Compared with L-1 and L-2, it controls the upper bound, therefore suppressing the influence of outliers. Compared with Logcosh and Huber, it not only adjusts the loss of small deviations but also suppresses the loss of large deviations. Compared with Wing loss, it considers the characteristics of data in RSs and therefore is more suitable to the matrix factorization model. Experiments are undertaken on six real-world recommendation datasets in comparison with six popular loss functions. Results show that Seagull loss function has superiority to all losses above on mean absolute error (MAE) and root mean square error (RMSE) and slight inferiority to L-2 and Wing on hit ratio (HR), mean average precision (MAP), and normalized discounted cumulative gain (NDCG).

关键词： Training loss measurement Recommender systems Fans Electronic mail Uncertainty Motion pictures Computer science Vectors Upper bound loss function matrix factorization outlier recommender system (RS)

来源：评论

学校读者我要写书评

暂无评论

FPDIoU loss: A loss function for efficient bounding box regression of rotated object detection

引用

IMAGE AND VISION COMPUTING 2025年 154卷

作者： Ma, Siliang Xu, Yong South China Univ Technol Inst Comp Sci & Engn 382Outer Ring East Rd Guangzhou 510006 Guangdong Peoples R China Pengcheng Lab 2Xingke 1st St Shenzhen 518000 Guangdong Peoples R China

Bounding box regression is one of the important steps of object detection. However, rotation detectors often involve amore complicated loss based on SkewIoU which is unfriendly to gradient-based training. Most of the existing loss functions for rotated object detection calculate the difference between two bounding boxes only focus on the deviation of area or each points distance (e.g., G h - 1 , G and G ). The calculation process of some loss functions is extremely complex (e.g. G ). In order to improve the efficiency and accuracy of bounding box regression for rotated object detection, we proposed a novel metric for arbitrary shapes comparison based on minimum points distance, which takes most of the factors from existing loss functions for rotated object detection into account, i.e., the overlap or nonoverlapping area, the central points distance and the rotation angle. We also proposed a loss function called G based on four points distance for accurate bounding box regression focusing on faster and high quality anchor boxes. In the experiments, loss has been applied to state-of-the-art rotated object detection (e.g., RTMDET, H2RBox) models training with three popular benchmarks of rotated object detection including DOTA, DIOR, HRSC2016 and two benchmarks of arbitrary orientation scene text detection including ICDAR 2017 RRC-MLT and ICDAR 2019 RRC-MLT, which achieves better performance than existing loss functions. The code is available at https://***/JacksonMa618/FPDIoU

关键词： Rotated object detection Bounding box regression loss function Minimum points distance

来源：评论

学校读者我要写书评

暂无评论

Improving Crack Detection Precision of Concrete Structures Using U-Net Architecture and Novel DBCE loss function

引用

IEEE ACCESS 2025年 13卷 20903-20922页

作者： Prasetyo, Andrew Purnama, I. Ketut Eddy Yuniarno, Eko Mulyanto Suprobo, Priyo Inst Teknol Sepuluh Nopember Dept Elect Engn Surabaya 60111 Indonesia Inst Teknol Sepuluh Nopember Dept Comp Engn Surabaya 60111 Indonesia Inst Teknol Sepuluh Nopember Dept Civil Engn Surabaya 60111 Indonesia

Monitoring the health of infrastructure is critical to maintaining the integrity of concrete construction. Conventional crack detection methods that rely on visual inspection and image processing often produce inconsistent results. U-Net, an architecture often used in image processing, has limitations in crack segmentation, especially regarding the loss function. The main challenge is class imbalance, as cracks usually occupy a much smaller area compared to the background in concrete images. To address this, we developed an automated technique utilizing the U-Net framework with a novel loss function that we named DBCE loss. Unlike the typical BCED we added the LogCoshDice function to get an improvement so that the resulting loss will be better and can improve the performance of the model. The performance of our method is rigorously evaluated by combining DeepCrack, GAPS, Crack500, and CrackForest datasets to illustrate that the model can detect cracks in various material conditions. DeepCrack represents cracks in concrete with minimal distress, while GAPS represents cracks in asphalt. Crack500 covers cracks with significant distress such as gravel, and CrackForest focuses on small cracks in *** proposed U-Net model achieved Maximum accuracies between 98.58% and 99.22%, Maximum Dice coefficients from 88.27% to 98.03%, and Maximum F1-scores up to 97.11%. The proposed method is 7.69% ahead of the largest comparison method (DICE loss) on Average IOU, while on Average Precission is leading 6.72% ahead of the largest comparison method (DICE). Then in terms of Recall our method is 8.84% superior to the largest comparison method (BCE). Sera in terms of Average F1-Score is 14.64% ahead of the largest comparison method (DICE). These results show that our method has surpassed conventional loss methods such as BCE, DICE and BCED. This research can facilitate a more reliable infrastructure health monitoring process and help with crack detection in the future.

关键词： Image segmentation Training Inspection Data augmentation Accuracy Visualization Surface cracks Monitoring Surface treatment Stress Crack segmentation U-Net loss function deep learning structural health monitoring

来源：评论

学校读者我要写书评

暂无评论

YOLOv5-RF: a deep learning method for tailings pond identification in high-resolution remote sensing images based on improved loss function

引用

BIG EARTH DATA 2025年第1期9卷 100-126页

作者： Zhang, Weiming Jiang, Wenliang Li, Qiang Luo, Yi Zhang, Heng Jiao, Qisong Li, Yongsheng Jiang, Hongbo Minist Emergency Management China Natl Inst Nat Hazards 1 Anningzhuang Rd Beijing 100085 Peoples R China Tengyuan Coal Co Ltd Ordos Inner Mongolia Peoples R China

Tailings ponds are critical facilities in the mining industry, and accurate monitoring and management of these ponds are of paramount importance. However, conventional object detection methodologies, including recent advancements, often face significant challenges in addressing the complexities inherent to tailings pond environments. This is particularly due to deficiencies in their loss function design, which can result in protracted convergence times and suboptimal performance when detecting smaller targets. In this study, we introduce an innovative loss function termed the Rapid Intersection over Union (RIoU) loss function, which incorporates a focal weight and is integrated into the YOLOv5 object detection framework to develop the YOLOv5-RF model. This approach aims to enhance both convergence speed and improve convergence accuracy in the tailings pond identification process by comprehensively addressing the specific challenges posed by complex environmental conditions, thereby enhancing the precision and robustness of tailings pond target detection. It integrates the concepts of the central triangle and the aspect ratio of the circumscribed rectangle, assigning specific weights and penalty terms to optimize the model's performance in object detection tasks. We validated the efficacy of YOLOv5-RF through simulation experiments and high-resolution remote sensing images of tailings ponds. The experimental results indicate that RIoU facilitates faster convergence rates. Specifically, YOLOv5-RF achieves accuracy and recall rates that are 2% and 2.1% higher than those of YOLOv5, respectively. Furthermore, it completes 120 iterations in 1.08 hours less time compared to its predecessor model while exhibiting an inference time that is 11.7 ms shorter than that for YOLOv5. These findings suggest that our model significantly enhances processing speed without compromising accuracy levels. This research offers novel technical approaches as well as theoretical support for mo

关键词： Object detection YOLOv5 loss function tailings ponds remote sensing identification

来源：评论

学校读者我要写书评

暂无评论

A Novel Proactive Fault Tolerance loss function for Crack Segmentation

引用

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS 2025年第5期26卷 6361-6378页

作者： Li, Bingchao Li, Zihao Zong, Jianping Wang, Huaichao Li, Nansha Li, Haifeng Civil Aviat Univ China Coll Comp Sci & Technol Tianjin 300300 Peoples R China Nankai Univ Coll Artificial Intelligence Tianjin 300071 Peoples R China

Optimizing the generalization performance of road surface crack models in practical applications represents a challenging task. Especially for thin and irregular cracks with random and expansive topologies, the loss functions used in current deep learning-based crack segmentation models are sensitive to single pixels, which tends to cause the model to overfit the training data, diminishing its generalization ability in real-world scenarios. Therefore, we take the loss function as a starting point and explore the introduction of proactive fault tolerance mechanisms into the training process of the crack segmentation model, which is called Proactive Fault Tolerance loss (PFT loss), to enhance the generalization capability of model in actual applications. Specifically, the PFT loss function establishes correlations between the segmentation prediction pixels and the corresponding labeled pixels within the neighborhood window using Markov Random Fields (MRFs). The correlation is used as a reference for predicting relative shifts in segmented pixels. Proactive Fault Tolerance is performed on the loss between labeling and prediction to achieve a more natural and adaptive training method for crack segmentation. Full experiments are conducted on five public crack datasets and one self-constructed dataset. The experimental results indicate that the model trained with PFT loss has better segmentation performance compared to other loss functions.

关键词： Fault tolerant systems Fault tolerance Image segmentation Training Sensitivity Roads Adaptation models Computational modeling Data models Airports Crack segmentation proactive fault tolerance generalization performance loss function

来源：评论

学校读者我要写书评

暂无评论

loss function Optimization Method and Unsupervised Extraction Approach D-DBSCAN for Improving the Moving Target Perception of 3D Imaging Sonar

引用

JOURNAL OF MARINE SCIENCE AND ENGINEERING 2025年第3期13卷 529-529页

作者： Yu, Jingfeng Huang, Aigen Sun, Zhongju Huang, Rui Huang, Gao Zhao, Qianchuan Chinese Acad Sci Ganjiang Innovat Acad Ganzhou 341000 Peoples R China Tsinghua Univ Dept Automat Beijing 100190 Peoples R China

Imaging sonar is a crucial tool for underwater visual perception. Compared to 2D sonar images, 3D sonar images offer superior spatial positioning capabilities, although the data acquisition cost is higher and lacks open source references for data annotation, target detection, and semantic segmentation. This paper utilizes 3D imaging sonar to collect underwater data from three types of targets with 1534 effective frames, including a tire, mannequin, and table, in Liquan Lake, Shanxi Province, China. Based on these data, this study focuses on three innovative aspects as follows: rapid underwater data annotation, loss function optimization, and unsupervised moving target extraction in water. For rapid data annotation, a batch annotation method combining human expertise and multi-frame superposition is proposed. This method automatically generates single-frame target detection boxes based on multi-frame joint segmentation, offering advantages in speed, cost, and accuracy. For loss function optimization, a density-based loss function is introduced to address the issue of overfitting in dense regions due to the uneven distribution of point cloud data. By assigning different weights to data points in different density regions, the model pays more attention to accurate predictions in a sparse area, resulting in a 6.939 improvement in mIOU for semantic segmentation tasks, while lakebed mIOU achieved a high score of 99.28. For unsupervised moving target extraction, a multi-frame joint unsupervised moving target association extraction method called the Double DBSCAN, D-DBSCAN, is proposed. This method simulates human visual sensitivity to moving targets in water and uses a joint D-DBSCAN spatial clustering approach with single-frame and inter-frame superposition, achieving an improvement of 21.3 points in mAP. Finally, the paper summarizes the three proposed innovations and provides directions for further research.

关键词： 3D imaging sonar annotation loss function D-DBSCAN target detection

来源：评论

学校读者我要写书评

暂无评论

loss function inversion for improved crack segmentation in steel bridges using a CNN framework

引用

AUTOMATION IN CONSTRUCTION 2025年 170卷

作者： Kompanets, Andrii Duits, Remco Pai, Gautam Leonetti, Davide Snijder, H. H. (Bert) Eindhoven Univ Technol Dept Built Environm Eindhoven Netherlands Eindhoven Univ Technol Eindhoven Artificial Intelligence Syst Inst Eindhoven Netherlands Eindhoven Univ Technol Dept Math & Comp Sci Eindhoven Netherlands

Automating bridge visual inspection using deep learning algorithms for crack detection in images is a prominent way to make these inspections more effective. This paper addresses several challenges associated with crack detection: (1) data imbalance, caused by a small crack area as compared to the background, and (2) a high false positive rate, due to a large amount of crack-like features in the background. First, anew benchmark dataset is presented, containing images of cracks in steel bridges along with pixel-wise annotations. Secondly, the importance of incorporating background patches is examined to assess their impact on network performance when applied to high resolution images of cracks in steel bridges. Finally, a loss function is introduced that enables the use of a relatively large number of background patches in neural network training. The proposed approaches yield a significant reduction in false positive rates, thereby improving the overall performance of crack segmentation.

关键词： Fatigue crack segmentation Computer vision loss function Steel bridge inspection Crack in steel bridge dataset

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：