检索结果-内蒙古大学图书馆

Multi-Scale Terahertz Image Reconstruction ⋆,⋆⋆

学校读者我要写书评

暂无评论

SSRN

SSRN 2024年

作者： Su, Zhipeng Zhang, Yixiong Zhou, Jianyang Shi, Jianghong Qi, Feng Zhang, Xiao-Ping Ji, Chunlin School of Informatics Xiamen University National Demonstrative Software School Xiamen University China School of Electronic Science and Engineering National Model Microelectronics College Xiamen University China Shenyang Institute of Automation Chinese Academy of Sciences China Department of Electrical Computer & Biomedical Engineering Finance Department The Ted Rogers School of Management Toronto Metropolitan University Canada Kuang-Chi Institute of Advanced Technology State Key Laboratory of Meta-RF Electromagnetic Modulation Technology Shen Zhen China

This paper proposes a novel terahertz (THz) image recovery algorithm and a new THz image dataset is publicly available. Because of transmission noise, artificial errors and problems with diffraction phenomena are among the main problems that cause degradation of THz images. The point spread function (PSF) in real transmission is first obtained using a terahertz imaging system. Experiments can generate degraded images and clean image datasets for supervised training of neural networks via terahertz PSF. A Depthwise Dense Instantiation Normalization Block (DwDIN Block) has been proposed to recover the datasets. We constructed a powerful multi-stage network (DwDINet) at each scale by utilizing the DwDIN module. For input and output, the results of each stage of the network and the original images are taken as input into the next layer in order to enhance the background information. DwDINet achieves state-of-the-art (SOTA) results on the two main THz datasets. To verify the generality of the model, we performed experiments on various image recovery tasks. According to the experimental results, the proposed model is of great potential for the field of low-level image clarification. © 2024, The Authors. All rights reserved.

关键词： Optical transfer function

Tackling Noisy labels with Network Parameter Additive Decomposition

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Wang, Jingyi Xia, Xiaobo Lan, Long Wu, Xinghao Yu, Jun Yang, Wenjing Han, Bo Liu, Tongliang The Department of Intelligent Data Science College of Computer Science and Technology National University of Defense Technology Changsha410073 China The Trustworthy Machine Learning Lab School of Computer Science Faculty of Engineering University of Sydney DarlingtonNSW2008 Australia The State Key Laboratory of Virtual Reality Technology and Systems School of Computer Science and Engineering Beihang University Beijing100191 China The Department of Automation University of Science and Technology of China Hefei230026 China The Department of Computer Science Hong Kong Baptist University Hong Kong

Given data with noisy labels, over-parameterized deep networks suffer overfitting mislabeled data, resulting in poor generalization. The memorization effect of deep networks shows that although the networks have the ability to memorize all noisy data, they would first memorize clean training data, and then gradually memorize mislabeled training data. A simple and effective method that exploits the memorization effect to combat noisy labels is early stopping. However, early stopping cannot distinguish the memorization of clean data and mislabeled data, resulting in the network still inevitably overfitting mislabeled data in the early training stage. In this paper, to decouple the memorization of clean data and mislabeled data, and further reduce the side effect of mislabeled data, we perform additive decomposition on network parameters. Namely, all parameters are additively decomposed into two groups, i.e. , parameters w are decomposed as w = σ + γ. Afterward, the parameters σ are considered to memorize clean data, while the parameters γ are considered to memorize mislabeled data. Benefiting from the memorization effect, the updates of the parameters σ are encouraged to fully memorize clean data in early training, and then discouraged with the increase of training epochs to reduce interference of mislabeled data. The updates of the parameters γ are the opposite. In testing, only the parameters σ are employed to enhance generalization. Extensive experiments on both simulated and real-world benchmarks confirm the superior performance of our method. The implementation is available at: https://***/wangjingyi9924/TNLPAD. Copyright © 2024, The Authors. All rights reserved.

关键词： Benchmarking

Learning the Relation between Similarity Loss and Clustering Loss in Self-Supervised Learning

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Ge, Jidong Liu, Yuxiang Gui, Jie Fang, Lanting Lin, Ming Kwok, James Tin-Yau Huang, LiGuo Luo, Bin State Key Laboratory for Novel Software Technology Software Institute Nanjing University Nanjing210093 China The School of Cyber Science and Engineering Southeast University Purple Mountain Laboratories Nanjing210000 China *** L.L.C. United States Alibaba Group China The Department of Computer Science and Engineering The Hong Kong University of Science and Technology 999077 Hong Kong The Department of Computer Science Southern Methodist University DallasTX75205 United States

Self-supervised learning enables networks to learn discriminative features from massive data itself. Most state-of-the-art methods maximize the similarity between two augmentations of one image based on contrastive learning. By utilizing the consistency of two augmentations, the burden of manual annotations can be freed. Contrastive learning exploits instance-level information to learn robust features. However, the learned information is probably confined to different views of the same instance. In this paper, we attempt to leverage the similarity between two distinct images to boost representation in self-supervised learning. In contrast to instance-level information, the similarity between two distinct images may provide more useful information. Besides, we analyze the relation between similarity loss and feature-level cross-entropy loss. These two losses are essential for most deep learning methods. However, the relation between these two losses is not clear. Similarity loss helps obtain instance-level representation, while feature-level cross-entropy loss helps mine the similarity between two distinct images. We provide theoretical analyses and experiments to show that a suitable combination of these two losses can get state-of-the-art results. Code is available at https://***/guijiejie/ICCL. Copyright © 2023, The Authors. All rights reserved.

关键词： Image classification

Why Does Little Robustness Help? A Further Step Towards Understanding Adversarial Transferability

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Zhang, Yechao Hu, Shengshan Zhang, Leo Yu Shi, Junyu Li, Minghui Liu, Xiaogeng Wan, Wei Jin, Hai School of Cyber Science and Engineering Huazhong University of Science and Technology China School of Software Engineering Huazhong University of Science and Technology China School of Information and Communication Technology Griffith University Australia School of Computer Science and Technology Huazhong University of Science and Technology China National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Hubei Key Laboratory of Distributed System Security China Hubei Engineering Research Center on Big Data Security China Cluster and Grid Computing Lab

Adversarial examples for deep neural networks (DNNs) have been shown to be transferable: examples that successfully fool one white-box surrogate model can also deceive other black-box models with different architectures. Although a bunch of empirical studies have provided guidance on generating highly transferable adversarial examples, many of these findings fail to be well explained and even lead to confusing or inconsistent advice for practical use. In this paper, we take a further step towards understanding adversarial transferability, with a particular focus on surrogate aspects. Starting from the intriguing "little robustness" phenomenon, where models adversarially trained with mildly perturbed adversarial samples can serve as better surrogates for transfer attacks, we attribute it to a trade-off between two dominant factors: model smoothness and gradient similarity. Our research focuses on their joint effects on transferability, rather than demonstrating the separate relationships alone. Through a combination of theoretical and empirical analyses, we hypothesize that the data distribution shift induced by off-manifold samples in adversarial training is the reason that impairs gradient similarity. Building on these insights, we further explore the impacts of prevalent data augmentation and gradient regularization on transferability and analyze how the trade-off manifest in various training methods, thus building a comprehensive blueprint for the regulation mechanisms behind transferability. Finally, we provide a general route for constructing superior surrogates to boost transferability, which optimizes both model smoothness and gradient similarity simultaneously, e.g., the combination of input gradient regularization and sharpness-aware minimization (SAM), validated by extensive experiments. In summary, we call for attention to the united impacts of these two factors for launching effective transfer attacks, rather than optimizing one while ignoring the other,

关键词： Deep neural networks

Reler: Relearning Controversial Regions to Accurately Segment Nasopharyngeal Carcinoma

学校读者我要写书评

暂无评论

Reler: Relearning Controversial Regions to Accurately Segmen...

2022 IEEE International Conference on Bioinformatics and Biomedicine, BIBM 2022

作者： Tao, Guihua Li, Haojiang Lu, Dandan Ling, Ziqin Liu, Lizhi Cai, Hongmin South China University of Technology School of Computer Science and Engineering Guangzhou China Sun Yat-sen University Cancer Center Sun Yat-sen University State Key Laboratory of Oncology in South China Collaborative Innovation Center for Cancer Medicine Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy Guangzhou China Pazhou Lab Brain and Affective Cognitive Research Center Guangzhou China

ISBN: (纸本)9781665468190

Accurate nasopharyngeal carcinoma (NPC) segmentation is significant in preventing local recurrence and improving patients' survival rates. However, existing deep learning-based methods often yield unsatisfactory segmentation results, especially in fine-grained detail. Because NPC is a tiny and infiltrative tumor with a huge background, traditional deep neural networks tend to be dominated by salient information, thus missing the fine-grained details of NPC. To achieve accurate NPC segmentation, a relearning controversial regions method (Reler) is proposed. It consists of three modules, including the controversial features generator (CFG), controversial features finding module (CFF), and controversial regions arbitration module (CRA). First, CFG constructs global and local feature extractors to generate two types of different features. Then, CFF finds the controversial features and corresponding regions by comparing the global and local features' estimates of the segmentation results of the same input regions. Next, the CRA focuses on controversial features, relearns new features, and produces new segmentation results through a proposed Transformer-based self-attention network. Finally, the uncontroversial segmentation results from CFF and CRA are combined as the final segmentation results. Extensive experiments are conducted on a large NPC dataset containing 6342 images from 596 patients. The experimental results show that the proposed method Reler is effective and superior to the nine state-of-the-art methods. © 2022 IEEE.

关键词： Deep neural networks

A Tensor Solution for Health Indicator Construction of Metro Wheelset Degradation with Irregular Noise

学校读者我要写书评

暂无评论

A Tensor Solution for Health Indicator Construction of Metro...

Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD), International Conference on

作者： Yu Wang Wentao Mao Linlin Kou keying Liu School of Computer and Information Engineering Henan Normal University Xinxiang China State Key Lab of Rail Traffic Control & Safety Beijing Jiaotong University Beijing China

During the operation of metro vehicles, wheel flange and wheel diameter continue to be abrased, so wheel degradation is inevitable. Influenced by the states of welded rail, subgrade and turnout, applied load, speed of running and other components, the degradation process of wheelset is vulnerable to irregular noise interference, which leads to abnormal fluctuation of vibration signal. As a result, the degradation process is hard to be accurately described. To address this problem, this paper proposes a new health indicator construction method for metro wheelset based on tensor reconstruction. First, tensor Tucker decomposition is utilized to obtain the core tensor of original signal, and then tensor reconstruction is applied to transform the signal into a new degradation sequence with noise reduction. Second, the Savitzky-Golay filter is employed to remove the irregular trend from the obtained degradation sequence. Finally, deep autoencoder network is used to extract deep degradation features, and after dimension reduction, a health indicator of wheelset degradation process can be obtained. The effectiveness of the proposed method is verified with Beijing subway wheelset degradation dataset in 2020. The results show that the constructed health indicator can accurately describe the whole degradation process with good trendability and monotonicity. More importantly, the proposed method possess good practicability since the key changing parts in the health indicator can correspond well to the actual maintenance records.

关键词： Degradation Vibrations Tensors Welding Wheels Interference Transforms

Prototype-Wise Self-Knowledge Distillation for Few-Shot Segmentation

学校读者我要写书评

暂无评论

SSRN

SSRN 2023年

作者： Chen, Yadang Xu, Xinyu Wei, Chenchen Yang, Zhi-Xin School of Computer and Software Nanjing University of Information Science and Technology Nanjing210044 China Engineering Research Center of Digital Forensics Ministry of Education Nanjing University of Information Science and Technology Nanjing210044 China The State Key Laboratory of Internet of Things for Smart City Department of Electromechanical Engineering University of Macau 999078 China

Few-shot segmentation was proposed to obtain segmentation results for a image with an unseen class by referring to a few labeled samples. However, due to the limited number of samples, many few-shot segmentation models suffer from poor generalization. Prototypical network-based few-shot segmentation still has issues with spatial inconsistency and prototype bias. Since the target class has different appearance in each image, some specific features in the prototypes generated from the support image and its mask do not accurately reflect the generalized features of the target class. To address the support prototype consistency issue, we put forward two modules: Data Augmentation Self-knowledge Distillation (DASKD) and Prototype-wise Regularization (PWR). The DASKD module focuses on enhancing spatial consistency by using data augmentation and self-knowledge distillation. Self-knowledge distillation helps the model acquire generalized features of the target class and learn hidden knowledge from the support images. The PWR module focuses on obtaining a more representative support prototype by conducting prototype-level loss to obtain support prototypes closer to the category center. Broad evaluation experiments on PASCAL-5i and COCO-20i demonstrate that our model outperforms the prior works on few-shot segmentation. Our approach surpasses the state of the art by 7.5% in PASCAL-5i and 4.2% in COCO-20i. © 2023, The Authors. All rights reserved.

关键词： Distillation

A Four-Pronged Defense Against Byzantine Attacks in Federated Learning

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Wan, Wei Hu, Shengshan Li, Minghui Lu, Jianrong Zhang, Longling Zhang, Leo Yu Jin, Hai School of Cyber Science and Engineering Huazhong University of Science and Technology China School of Software Engineering Huazhong University of Science and Technology China School of Information and Communication Technology Griffith University Australia School of Computer Science and Technology Huazhong University of Science and Technology China National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Hubei Key Laboratory of Distributed System Security China Hubei Engineering Research Center on Big Data Security China Cluster and Grid Computing Lab

Federated learning (FL) is a nascent distributed learning paradigm to train a shared global model without violating users' privacy. FL has been shown to be vulnerable to various Byzantine attacks, where malicious participants could independently or collusively upload well-crafted updates to deteriorate the performance of the global model. However, existing defenses could only mitigate part of Byzantine attacks, without providing an all-sided shield for FL. It is difficult to simply combine them as they rely on totally contradictory assumptions. In this paper, we propose FPD, a four-pronged defense against both non-colluding and colluding Byzantine attacks. Our main idea is to utilize absolute similarity to filter updates rather than relative similarity used in existingI works. To this end, we first propose a reliable client selection strategy to prevent the majority of threats in the bud. Then we design a simple but effective score-based detection method to mitigate colluding attacks. Third, we construct an enhanced spectral-based outlier detector to accurately discard abnormal updates when the training data is not independent and identically distributed (non-IID). Finally, we design update denoising to rectify the direction of the slightly noisy but harmful updates. The four sequentially combined modules can effectively reconcile the contradiction in addressing non-colluding and colluding Byzantine attacks. Extensive experiments over three benchmark image classification datasets against four state-of-the-art Byzantine attacks demonstrate that FPD drastically outperforms existing defenses in IID and non-IID scenarios (with 30% improvement on model accuracy). © 2023, CC BY.

关键词： Classification (of information)

PAC Learning of Deterministic One-Clock Timed Automata 22nd

学校读者我要写书评

暂无评论

PAC Learning of Deterministic One-Clock Timed Automata

22nd International Conference on Formal engineering Methods, ICFEM 2020

作者： Shen, Wei An, Jie Zhan, Bohua Zhang, Miaomiao Xue, Bai Zhan, Naijun School of Software Engineering Tongji University Shanghai China State Key Laboratory of Computer Science Institute of Software CAS Beijing China University of Chinese Academy of Sciences Beijing China

ISBN: (纸本)9783030634056

We study the problem of learning deterministic one-clock timed automata in the framework of PAC (probably approximately correct) learning. The use of PAC learning relaxes the assumption of having a teacher that can answer equivalence queries exactly, replacing it with approximate answers from testing on a set of samples. The framework provides correctness guarantees in terms of error and confidence parameters. We further discuss several improvements to the basic PAC algorithm. This includes a special sampling method, and the use of comparator and counterexample minimization to reduce the number of equivalence queries. We implemented a prototype for our learning algorithm, and conducted experiments on learning the TCP protocol as well as a number of randomly generated automata. The results demonstrate the effectiveness of our approach, as well as the importance of the various improvements for learning complex models. © 2020, Springer Nature Switzerland AG.

关键词： Clocks