检索结果-内蒙古大学图书馆

2024 Conference on Empirical Methods in Natural Language processing, EMNLP 2024

作者： Zhang, Mozhi Huang, Mianqiu Shi, Rundong Guo, Linsen Peng, Chong Yan, Peng Zhou, Yaqian Qiu, Xipeng School of Computer Science Fudan University China Shanghai Key Laboratory of Intelligent Information Processing Fudan University China Meituan China

ISBN: (纸本)9798891761643

Large language models optimized with techniques like RLHF have achieved good alignment in being helpful and harmless. However, post-alignment, these language models often exhibit overconfidence, where the expressed confidence does not accurately calibrate with their correctness rate. In this paper, we decompose the language model confidence into the Uncertainty about the question and the Fidelity to the answer generated by language models. Then, we propose a plug-and-play method, UF Calibration, to estimate the confidence of language models. Our method has shown good calibration performance by conducting experiments with 6 RLHF-LMs on four MCQA datasets. Moreover, we propose two novel metrics, IPR and CE, to evaluate the calibration of the model, and we have conducted a detailed discussion on Truly Well-Calibrated Confidence for large language models. Our method could serve as a strong baseline, and we hope that this work will provide some insights into the model confidence calibration. © 2024 Association for Computational Linguistics.

关键词： Calibration

来源：评论

学校读者我要写书评

暂无评论

Unsigned Road Incidents Detection Using Improved RESNET From Driver-View Images

IEEE Transactions on Artificial Intelligence

引用

IEEE Transactions on Artificial Intelligence 2025年第5期6卷 1203-1216页

作者： Li, Changping Wang, Bingshu Zheng, Jiangbin Zhang, Yongjun Chen, C.L. Philip Northwestern Polytechnical University School of Software Xi’an710129 China Shenzhen University Guangdong Provincial Key Laboratory of Intelligent Information Processing Shenzhen Key Laboratory of Media Security Shenzhen518060 China Guizhou University State Key Laboratory of Public Big Data College of Computer Science and Technology Guiyang550025 China South China University of Technology School of Computer Science and Engineering Guangzhou510641 China

Frequent road incidents cause significant physical harm and economic losses globally. The key to ensuring road safety lies in accurately perceiving surrounding road incidents. However, the highly dynamic nature of traffic introduces significant challenges, particularly in detecting sudden and temporary incidents. In this article, we propose a novel detection framework, multihead attention ResNet with dynamic bottleneck (DB_RESNET_MHA), to identify physical unsigned road incidents. Our approach introduces three key innovations. First, we develop a tailored data augmentation strategy to create images that closely mimic the complex variations found in real-world road environments. Second, we enhance model expressiveness by employing attention mechanisms to nonlinearly integrate convolutional kernels within the residual network. Furthermore, we refine the prediction head by applying spatially distinct attention weights, enabling the model to capture intricate correlations between different features more effectively. To demonstrate the effectiveness of our method, we create a dataset for unsigned road incidents (UnsignRI), comprising a total of 16 323 images that capture 12 distinct types of incidents. It stands out as the most comprehensive dataset in the field, encompassing a wide range of geographical features and incident categories. Experimental results show that DB_RESNET_MHA achieves an average accuracy of 96.2% and a f1-score of 0.955 across various categories of unsigned incidents, surpassing other models. © 2020 IEEE.

关键词： Highway accidents

来源：评论

学校读者我要写书评

暂无评论

Improving Transferability Reversible Adversarial Examples Based on Flipping Transformation 9th

Improving Transferability Reversible Adversarial Examples Ba...

引用

9th International Conference of Pioneering computer Scientists, Engineers and Educators, ICPCSEE 2023

作者： Fang, Youqing Jia, Jingwen Yang, Yuhai Lyu, Wanli Key Lab of Intelligent Computing and Signal Processing of Ministry of Education School of Computer Science and Technology Anhui University Hefei230601 China

ISBN: (纸本)9789819959679

Adding subtle perturbations to an image can cause the classification model to misclassify, and such images are called adversarial examples. Adversarial examples threaten the safe use of deep neural networks, but when combined with reversible data hiding (RDH) technology, they can protect images from being correctly identified by unauthorized models and recover the image lossless under authorized models. Based on this, the reversible adversarial example (RAE) is rising. However, existing RAE technology focuses on feasibility, attack success rate and image quality, but ignores transferability and time complexity. In this paper, we optimize the data hiding structure and combine data augmentation technology, which flips the input image in probability to avoid overfitting phenomenon on the dataset. On the premise of maintaining a high success rate of white-box attacks and the image’s visual quality, the proposed method improves the transferability of reversible adversarial examples by approximately 16% and reduces the computational cost by approximately 43% compared to the state-of-the-art method. In addition, the appropriate flip probability can be selected for different application scenarios. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

An Amalgamated CNN-Transformer Network for Lightweight Image Super-Resolution

Journal of Network Intelligence

引用

Journal of Network Intelligence 2024年第3期9卷 1376-1387页

作者： Fang, Jinsheng Lin, Hanjiang Zeng, Kun School of Computer Science and Engineering Minnan Normal University Zhangzhou363000 China Key Laboratory of Data Science and Intelligence Application Fujian Province University Fujian Zhangzhou363000 China School of Computer and Control engineering Fujian Provincial Key Laboratory of Information Processing and Intelligent Control Minjiang University Fuzhou350108 China

Recently, Transformer-based methods for single image super-resolution (SISR) have achieved better performance advantages than the methods based on convolutional neural network (CNN). Exploiting self-attention mechanism to model global context definitely improves the SR results. However, the neglect of local information will bring inevitable reduction of the network performance. In this work, we propose an Amalgamated CNN-Transformer network for lightweight SR, namely ACTSR. Specifically, an amalgamated CNN-Transformer block (ACTB) is developed to extract the useful information of both local and global features. By employing stacked ACTBs, our ACTSR extracts more informative deep features beneficially for super-resolution reconstruction to improve network performance while keeps lightweight and flexible enough. Extensive experiments on commonly used benchmark datasets validate our ACTSR outperforms the advanced competitors. Our codes are available at: https://***/ginsengf/ACTSR. © 2024, Taiwan Ubiquitous Information CO LTD. All rights reserved.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Multi-speaker Emotional Speech Synthesis Based on Self-Supervised Speech Representations 5

Multi-speaker Emotional Speech Synthesis Based on Self-Super...

引用

5th IEEE International Conference on Pattern Recognition and Machine Learning, PRML 2024

作者： Yin, Mingming Huang, Zhihua Lu, Kexin Zhou, Haiyang School of Computer Science and Technology Xinjiang University Urumqi China School of Computer Science and Technology Xinjiang University Key Laboratory of Signal Detection and Processing Urumqi China

ISBN: (纸本)9798350355925

In recent years, emotional speech synthesis has shown considerable progress. However, some existing emotional speech synthesis methods only model emotion from a single scale, resulting in only global or average emotional expression, unable to synthesize significant emotional speech. In this paper, we propose a multi-scale multi-speaker emotional speech synthesis method based on self-supervised speech model HuBERT. Firstly, the emotion information is extracted from the feature extractor HuBERT. Secondly, the extracted emotion information is modeled by coarse-grained emotion modeling and fine-grained emotion modeling respectively to obtain utterance-level emotion embedding and phoneme-level emotion embedding. Through subjective and objective evaluation, we verify that the method is significantly better than the comparison model in terms of speech quality and emotional expression. © 2024 IEEE.

关键词： emotional speech synthesis self-supervised speech representation speech synthesis

来源：评论

学校读者我要写书评

暂无评论

Enhancing defocus blur in three-dimensional scenes via high-speed optical axial scanning and deep learning 2

Enhancing defocus blur in three-dimensional scenes via high-...

引用

2nd International Academic Conference on Optics and Photonics, IACOP 2024

作者： Li, Zilong Tang, Chenxi Dong, Jiaqing Liu, Xuan Lin, Jiabin Liu, Qiegen Song, Xianlin School of Information Engineering Nanchang University Nanchang330031 China Jiangxi Provincial Key Laboratory of Advanced Signal Processing and Intelligent Communications Nanchang University Nanchang330031 China

ISBN: (数字)9781510688827

ISBN: (纸本)9781510688810

Mitigating image defocus blur induced by the limited depth of field in imaging systems and reconstructing the three-dimensional depth of scenes represents a significant hurdle. In this paper, a method for enhancing defocus blur in three-dimensional scenes via high-speed optical axial scanning and deep learning was proposed. The optical axial scanning system functions by synchronizing a liquid electronically tunable lens and a CCD camera to capture multi-focus image sequences. The system efficiently captures real-world images with a fast acquisition time of ~50 ms per image. The multi-focus image sequences are then fused to all-in-focus images. The multi-focus images and all-in-focus images (ground truth) serve as the dataset for the deep learning neural network. Defocused images can be reconstructed using the trained deep learning network. Furthermore, by establishing a direct relationship between the focal depth and the driving current applied to the liquid electronically tunable lens, the three-dimensional depth of the scene can be rapidly determined. Experimental results validate the feasibility and effectiveness of this method. It is expected to expand the applications of image defocus enhancement in fields such as medical imaging, 3D shape measurement, and augmented reality. © 2025 SPIE.

关键词： CCD cameras

来源：评论

学校读者我要写书评

暂无评论

Adversarial Example Generation Method Based on Probability Histogram Equalization 42

Adversarial Example Generation Method Based on Probability H...

引用

42nd Chinese Control Conference, CCC 2023

作者： Fang, Youqing Jia, Jingwen Yang, Yuhai Lyu, Wan-Li School of Computer Science and Technology Key Lab of Intelligent Computing and Signal Processing of Ministry of Education Anhui University Hefei230601 China

ISBN: (纸本)9789887581543

CNNs (Convolutional Neural Networks) have a good performance on most classification tasks, but they are vulnerable when meeting adversarial examples. Research and design of highly aggressive adversarial examples can help enhance the security and robustness of CNNs. The transferability of adversarial examples is still low in black-box settings. Therefore, an adversarial example method based on probability histogram equalization, namely HE-MI-FGSM (Histogram Equalization Momentum Iterative Fast Gradient Sign Method) is proposed. In each iteration of the adversarial example generation process, the original input image is randomly histogram equalized, and then the gradient is calculated to generate adversarial perturbations to mitigate overfitting in the adversarial example. The effectiveness of the method is verified on the ImageNet dataset. Compared with the advanced method I-FGSM (Iterative Fast Gradient Sign Method) and MI-FGSM (Momentum I-FGSM), the attack success rate in the adversarial training network increased by 27.9% and 7.7% on average, respectively. © 2023 Technical Committee on Control Theory, Chinese Association of Automation.

关键词： Iterative methods

来源：评论

学校读者我要写书评

暂无评论

Improving BERT Fine-Tuning via Self-Ensemble and Self-Distillation

引用

Journal of computer Science & Technology 2023年第4期38卷 853-866页

作者：许一格邱锡鹏周浬皋黄萱菁 School of Computer Science Fudan UniversityShanghai 200433China Shanghai Key Laboratory of Intelligent Information Processing Fudan UniversityShanghai 200433China Huawei Technologies Co. Ltd.Hangzhou 310052China

Fine-tuning pre-trained language models like BERT have become an effective way in natural language processing(NLP)and yield state-of-the-art results on many downstream *** studies on adapting BERT to new tasks mainly focus on modifying the model structure,re-designing the pre-training tasks,and leveraging external data and *** fine-tuning strategy itself has yet to be fully *** this paper,we improve the fine-tuning of BERT with two effective mechanisms:self-ensemble and *** self-ensemble mechanism utilizes the checkpoints from an experience pool to integrate the teacher *** order to transfer knowledge from the teacher model to the student model efficiently,we further use knowledge distillation,which is called self-distillation because the distillation comes from the model itself through the time *** on the GLUE benchmark and the Text Classification benchmark show that our proposed approach can significantly improve the adaption of BERT without any external data or *** conduct exhaustive experiments to investigate the efficiency of the self-ensemble and self-distillation mechanisms,and our proposed approach achieves a new state-of-the-art result on the SNLI dataset.

关键词： BERT deep learning fine-tuning natural language processing(NLP) pre-training model

来源：评论

学校读者我要写书评

暂无评论

An Efficient Multi-objective Approach Based on Golden Jackal Search for Dynamic Economic Emission Dispatch

引用

Journal of Bionic Engineering 2024年第3期21卷 1541-1566页

作者： keyu Zhong Fen Xiao Xieping Gao Key Laboratory of Computing and Stochastic Mathematics of Ministry of Education Hunan Normal UniversityChangsha410081China School of Mathematics and Statistics Hunan Normal UniversityChangsha410081China School of Computer Science Xiangtan UniversityXiangtan411105China Hunan Provincial Key Laboratory of Intelligent Computing and Language Information Processing Hunan Normal UniversityChangsha410081China

Dynamic Economic Emission Dispatch(DEED)aims to optimize control over fuel cost and pollution emission,two conflicting objectives,by scheduling the output power of various units at specific *** many methods well-performed on the DEED problem,most of them fail to achieve expected results in practice due to a lack of effective trade-off mechanisms between the convergence and diversity of non-dominated optimal dispatching *** address this issue,a new multi-objective solver called Multi-Objective Golden Jackal Optimization(MOGJO)algorithm is proposed to cope with the DEED *** proposed algorithm first stores non-dominated optimal solutions found so far into an ***,it chooses the best dispatching solution from the archive as the leader through a selection mechanism designed based on elite selection strategy and Euclidean distance index *** mechanism can guide the algorithm to search for better dispatching solutions in the direction of reducing fuel costs and pollutant ***,the basic golden jackal optimization algorithm has the drawback of insufficient search,which hinders its ability to effectively discover more Pareto *** this end,a non-linear control parameter based on the cosine function is introduced to enhance global exploration of the dispatching space,thus improving the efficiency of finding the optimal dispatching *** proposed MOGJO is evaluated on the latest CEC benchmark test functions,and its superiority over the state-of-the-art multi-objective optimizers is highlighted by performance ***,empirical results on 5-unit,10-unit,IEEE 30-bus,and 30-unit systems show that the MOGJO can provide competitive compromise scheduling solutions compared to published DEED ***,in the analysis of the Pareto dominance relationship and the Euclidean distance index,the optimal dispatching solutions provided by MOGJO are the closest to the ideal solutions for minimizing fuel costs

关键词： Dynamic economic emission dispatch Multi-objective optimization Golden jackal Euclidean distance index

来源：评论

学校读者我要写书评

暂无评论

APFA:Ameliorated Pathfinder Algorithm for Engineering Applications

引用

Journal of Bionic Engineering 2024年第3期21卷 1592-1616页

Pathfinder algorithm(PFA)is a swarm intelligent optimization algorithm inspired by the collective activity behavior of swarm animals,imitating the leader in the population to guide followers in finding the best food *** algorithm has the characteristics of a simple structure and high ***,PFA faces challenges such as insufficient population diversity and susceptibility to local optima due to its inability to effectively balance the exploration and exploitation *** paper proposes an Ameliorated Pathfinder Algorithm called APFA to solve complex engineering optimization ***,a guidance mechanism based on multiple elite individuals is presented to enhance the global search capability of the ***,to improve the exploration efficiency of the algorithm,the Logistic chaos mapping is introduced to help the algorithm find more high-quality potential solutions while avoiding the worst ***,a comprehensive following strategy is designed to avoid the algorithm falling into local optima and further improve the convergence *** three strategies achieve an effective balance between exploration and exploitation overall,thus improving the optimization performance of the *** performance evaluation,APFA is validated by the CEC2022 benchmark test set and five engineering optimization problems,and compared with the state-of-the-art metaheuristic *** numerical experimental results demonstrated the superiority of APFA.

关键词： Pathfinder algorithm Swarm intelligent Metaheuristic Engineering problems

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：