检索结果-内蒙古大学图书馆

Label-independent hyperparameter-free self-supervised single-view deep subspace clustering

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Sindičić, Lovro Kopriva, Ivica Division of Computing and Data Science Ruđer Bošković Institute Bijenička cesta 54 Zagreb10000 Croatia

Deep subspace clustering (DSC) algorithms face several challenges that hinder their widespread adoption across various application domains. First, clustering quality is typically assessed using only the encoder’s output layer, disregarding valuable information present in intermediate layers. Second, most DSC approaches treat representation learning and subspace clustering as independent tasks, limiting their effectiveness. Third, they assume the availability of a held-out dataset for hyperparameter tuning, which is often impractical in real-world scenarios. Fourth, learning termination is commonly based on clustering error monitoring, requiring external labels. Finally, their performance often depends on post-processing techniques that rely on labeled data. To address these limitations, we introduce a novel single-view DSC approach that (i) minimizes a layer-wise self-expression loss using a joint representation matrix;(ii) optimizes a subspace-structured norm to enhance clustering quality;(iii) employs a multi-stage sequential learning framework, consisting of pre-training and fine-tuning, enabling the use of multiple regularization terms without hyperparameter tuning;(iv) incorporates a relative error-based self-stopping mechanism to terminate training without labels;and (v) retains a fixed number of leading coefficients in the learned representation matrix based on prior knowledge. We evaluate the proposed method on six datasets representing faces, digits, and objects. The results show that our method outperforms most linear SC algorithms with carefully tuned hyperparameters while maintaining competitive performance with the best-performing linear *** Codes 68 © 2025, CC BY.

关键词： Self-supervised learning

FLARE: Towards Universal dataset Purification against Backdoor Attacks

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Hou, Linshan Luo, Wei Hua, Zhongyun Chen, Songhua Yu Zhang, Leo Li, Yiming School of Computer Science and Technology Harbin Institute of Technology Guangdong Shenzhen518055 China School of Information Technology Deakin University Australia Guangdong Provincial Key Laboratory of Novel Security Intelligence Technologies Guangdong Shenzhen518055 China School of Information and Communication Technology Griffith University Southport Gold CoastQLD4215 Australia College of Computing and Data Science Nanyang Technological University 639798 Singapore

Deep neural networks (DNNs) are susceptible to backdoor attacks, where adversaries poison datasets with adversary-specified triggers to implant hidden backdoors, enabling malicious manipulation of model predictions. dataset purification serves as a proactive defense by removing malicious training samples to prevent backdoor injection at its source. We first reveal that the current advanced purification methods rely on a latent assumption that the backdoor connections between triggers and target labels in backdoor attacks are simpler to learn than the benign features. We demonstrate that this assumption, however, does not always hold, especially in all-to-all (A2A) and untargeted (UT) attacks. As a result, purification methods that analyze the separation between the poisoned and benign samples in the input-output space or the final hidden layer space are less effective. We observe that this separability is not confined to a single layer but varies across different hidden layers. Motivated by this understanding, we propose FLARE, a universal purification method to counter various backdoor attacks. FLARE aggregates abnormal activations from all hidden layers to construct representations for clustering. To enhance separation, FLARE develops an adaptive subspace selection algorithm to isolate the optimal space for dividing an entire dataset into two clusters. FLARE assesses the stability of each cluster and identifies the cluster with higher stability as poisoned. Extensive evaluations on benchmark datasets demonstrate the effectiveness of FLARE against 22 representative backdoor attacks, including all-to-one (A2O), all-to-all (A2A), and untargeted (UT) attacks, and its robustness to adaptive attacks. Copyright © 2024, The Authors. All rights reserved.

关键词： Deep neural networks

Convolutional neural network-assisted recognition of nanoscale L1_(2) ordered structures in face-centred cubic alloys

学校读者我要写书评

暂无评论

npj Computational Materials 2021年第1期7卷 60-68页

作者： Yue Li Xuyang Zhou Timoteo Colnaghi Ye Wei Andreas Marek Hongxiang Li Stefan Bauer Markus Rampp Leigh T.Stephenson Max-Planck Institut für Eisenforschung GmbH Max-Planck-Straße 140237 DüsseldorfGermany Max Planck Computing and Data Facility Gießenbachstraße 285748 GarchingGermany State Key Laboratory for Advanced Metals and Materials University of Science and Technology Beijing100083 BeijingChina Max Planck Institute for Intelligent Systems Max-Planck-Ring 472076 TübingenGermany

Nanoscale L12-type ordered structures are widely used in face-centered cubic(FCC)alloys to exploit their hardening capacity and thereby improve mechanical *** fine-scale particles are typically fully coherent with matrix with the same atomic configuration disregarding chemical species,which makes them challenging to be *** distribution maps(SDMs)are used to probe local order by interrogating the three-dimensional(3D)distribution of atoms within reconstructed atom probe tomography(APT)***,it is almost impossible to manually analyze the complete point cloud(>10 million)in search for the partial crystallographic information retained within the ***,we proposed an intelligent L1_(2)-ordered structure recognition method based on convolutional neural networks(CNNs).The SDMs of a simulated L1_(2)-ordered structure and the FCC matrix were firstly *** simulated images combined with a small amount of experimental data were used to train a CNN-based L1_(2)-ordered structure recognition ***,the approach was successfully applied to reveal the 3D distribution of L1_(2)–typeδ′–Al3(LiMg)nanoparticles with an average radius of 2.54 nm in a FCC Al-Li-Mg *** minimum radius of detectable nanodomain is even down to 5Å.The proposed CNN-APT method is promising to be extended to recognize other nanoscale ordered structures and even more-challenging short-range ordered phenomena in the near future.

关键词： alloys ordered cubic

Provably Secure Efficient Key-Exchange Protocol for Intelligent Supply Line Surveillance in Smart Grids

学校读者我要写书评

暂无评论

IEEE Internet of Things Journal 2025年第12期12卷 21784-21792页

作者： Ayub, Muhammad Faizan Li, Xiong Mahmood, Khalid Alenazi, Mohammed J.F. Das, Ashok Kumar Wang, Guijuan University of Electronic Science and Technology of China School of Computer Science and Engineering Sichuan Chengdu12599 China National Yunlin University of Science and Technology Graduate School of Intelligent Data Science Yunlin Douliu64002 Taiwan Department of Computer Engineering Riyadh11451 Saudi Arabia International Institute of Information Technology Center for Security Theory and Algorithmic Research Hyderabad500 032 India Korea University College of Informatics Department of Computer Science and Engineering Anam-ro Seoul145 Korea Republic of Ministry of Education Shandong Computer Science Center Qilu University of Technology Shandong Academy of Sciences Key Laboratory of Computing Power Network and Information Security Jinan250353 China Shandong Provincial Key Laboratory of Industrial Network and Information System Security China Shandong Fundamental Research Center for Computer Science Jinan 250353 China

Intelligent supply line surveillance is critical for modern smart grids. Smart sensors and gateway nodes are strategically deployed along supply lines to achieve intelligent surveillance. They collect data continuously and transmit it to the control centre in real-time. It enables real-time monitoring, fault detection, and efficient energy management across distribution networks. This advanced surveillance system ensures continuous monitoring of supply lines, detecting anomalies, and optimizing operations to maintain the stability and reliability of the smart grid. However, the reliance of all participating nodes on public communication channels to transmit supply line surveillance data exposes these systems to critical cyber attacks. These cyber attacks include impersonation, physical tampering, ephemeral secret leakage, and desynchronization attacks. To address these issues, existing key-exchange protocols often fail to ensure robust security while imposing high computation and communication overheads, limiting their practicality for resource-constrained environments. Therefore, we propose a secure and efficient key exchange and tamper-resistant authentication protocol using Elliptic Curve Cryptography (ECC) and Physical Unclonable Functions (PUFs). The PUF mechanism provides robust resistance against physical tampering and cloning attacks, ensuring enhanced physical security for supply line devices. We validate the security robustness of the devised protocol using the Random or Real (ROR) model. Furthermore, informal security analysis demonstrates the protocol's robustness in rigorously resisting various attacks, including physical tampering, impersonation, ephemeral secret leakage and desynchronization. Moreover, we determine the performance evaluation that reveals the protocol's superior efficiency compared to competing protocols, achieving significant reductions of 28.64% in computation overhead and 9.96% in communication overhead. © 2014 IEEE.

关键词： Authentication Protocol

Optimization Study of Renal CT Image Classification Using Convolution Neural Networks

学校读者我要写书评

暂无评论

Optimization Study of Renal CT Image Classification Using Co...

International Conference on computing and Networking Technology (ICCNT)

作者： Afnaan. K K. V Nagaraja Tripty Singh Bhanu Prakash KN Department of Computer Science and Engineering Amrita School of Computing Bengaluru Amrita Vishwa Vidyapeetham India Department of Mathematics Amrita School of Engineering Bengaluru Amrita Vishwa Vidyapeetham India Principal Investigator Research Scientist Clinical Data Analytics and Radiomics Bioinformatic Institute (BII) A*STAR Singapore

Computer tomography (CT), also known as computed tomography, is a medical imaging method that generates detailed and precise horizontal or axial images of targeted regions of the body for diagnostic purposes. In this particular study, the focus was on performing image classification of CT scan images of the Kidney. This organ plays an essential role in detoxification, fluid balance, and maintaining electrolyte levels in the human body. The dataset used for the study contained 12,446 images into four labeled classes - Cyst, Tumor, Stone, and Normal. Convolution Neural Network (CNN) model is built for binary (Normal and Tumor) and multi-image Classification to classify the Renal CT images into four classes. CNN uses ReLU as an activation function for classifying these images. The Model is evaluated and compared based on Accuracy and Loss performance metrics by compiling the CNN model with different deep learning TensorFlow Keras optimizers for Version 2.11.0 for Binary Image Classification at 10,20, and 50 epochs. The CNN model is assembled with Adam Optimizer and Follows the Regularized Leader (FTRL) Optimizer. The Testing and training Accuracy and Losses achieved are evaluated with a learning rate of 0.001 and 0.0001. Adam Optimizer outperforms to be the best Optimizer for Binary and Multi-Image Classification.

关键词：

SAGIPS: A Scalable Asynchronous Generative Inverse Problem Solver

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Lersch, Daniel Schram, Malachi Dai, Zhenyu Rajput, Kishansingh Wu, Xingfu Sato, N. Childers, J. Taylor Data Science Department Thomas Jefferson National Accelerator Facility Newport NewsVA23606 United States Mathematics and Computer Science Division Argonne National Laboratory LemontIL60439 United States Theory Center Thomas Jefferson National Accelerator Facility Newport NewsVA23606 United States Leadership Computing Division Argonne National Laboratory LemontIL60439 United States

Large scale, inverse problem solving deep learning algorithms have become an essential part of modern research and industrial applications. The complexity of the underlying inverse problem often poses challenges to the algorithm and requires the proper utilization of high-performance computing systems. Most deep learning algorithms require, due to their design, custom parallelization techniques in order to be resource efficient while showing a reasonable convergence. In this paper we introduces a Scalable Asynchronous Generative workflow for solving Inverse Problems Solver (SAGIPS) on high-performance computing systems. We present a workflow that utilizes a parallelization approach where the gradients of the generator network are updated in an asynchronous ring-all-reduce fashion. Experiments with a scientific proxy application demonstrate that SAGIPS shows near linear weak scaling, together with a convergence quality that is comparable to traditional methods. The approach presented here allows leveraging GANs across multiple GPUs, promising advancements in solving complex inverse problems at scale. Copyright © 2024, The Authors. All rights reserved.

关键词： Inverse problems

Event Coreference Resolution with their Paraphrases and Argument-aware Embeddings 28

学校读者我要写书评

暂无评论

Event Coreference Resolution with their Paraphrases and Argu...

28th International Conference on Computational Linguistics, COLING 2020

作者： Zeng, Yutao Jin, Xiaolong Guan, Saiping Guo, Jiafeng Cheng, Xueqi School of Computer Science and Technology University of Chinese Academy of Sciences CAS Key Laboratory of Network Data Science and Technology Institute of Computing Technology Chinese Academy of Sciences China

ISBN: (纸本)9781952148279

Event coreference resolution aims to classify all event mentions that refer to the same real-world event into the same group, which is necessary to information aggregation and many downstream applications. To resolve event coreference, existing methods usually calculate the similarities between event mentions and between specific kinds of event arguments. However, they fail to accurately identify paraphrase relations between events and may suffer from error propagation while extracting event components (i.e., event mentions and their arguments). Therefore, we propose a new model based on Event-specific Paraphrases and Argument-aware Semantic Embeddings, thus called EPASE, for event coreference resolution. EPASE recognizes deep paraphrase relations in an event-specific context of sentences and can cover event paraphrases of more situations, bringing about a better generalization. Additionally, the embeddings of argument roles are encoded into event embedding without relying on a fixed number and type of arguments, which results in the better scalability of EPASE. Experiments on both within- and cross-document event coreference demonstrate its consistent and significant superiority compared to existing methods. © 2020 COLING 2020 - 28th International Conference on Computational Linguistics, Proceedings of the Conference. All rights reserved.

关键词： Embeddings

A new computational model for quantifying blood flow dynamics across myogenically-active cerebral arterial networks

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Coccarelli, Alberto Polydoros, Ioannis Drysdale, Alex Harraz, Osama F. Kadapa, Chennakesava Zienkiewicz Institute for Modelling Data and AI Faculty of Science and Engineering Swansea University Swansea United Kingdom Department of Mechanical Engineering Faculty of Science and Engineering Swansea University Swansea United Kingdom Department of Pharmacology Larner College of Medicine Vermont Center for Cardiovascular and Brain Health University of Vermont Burlington United States School of Computing Engineering and the Built Environment Edinburgh Napier University Edinburgh United Kingdom

Cerebral autoregulation plays a key physiological role by limiting blood flow changes in the face of pressure fluctuations. Although the involved cellular processes are mechanically driven, the quantification of haemodynamic forces in in-vivo settings remains extremely difficult and uncertain. In this work, we propose a novel computational framework for evaluating the blood flow dynamics across networks of myogenically active cerebral arteries, which can modulate their muscular tone to stabilize flow (and perfusion pressure) as well as to limit vascular intramural stress. The introduced framework is built on contractile (myogenically active) vascular wall mechanics and blood flow dynamics models, which can be numerically coupled in either a weak or strong way. We investigate the time dependency of the vascular wall response to pressure changes at both single vessel and network levels. The robustness of the model was assessed by considering different types of inlet signals and numerical settings in an idealized vascular network formed by a middle cerebral artery and its three generations. For the vessel size and boundary conditions considered, weak coupling ensured accurate results with a lower computational cost. To complete the analysis, we evaluated the effect of an upstream pressure surge on the haemodynamics of the vascular network. This provided a clear quantitative picture of how pressure and flow are redistributed across each vessel generation upon inlet pressure changes. This work paves the way for future combined experimental-computational studies aiming to decipher cerebral autoregulation. © 2024, CC BY.

关键词： Hemodynamics

Energy Efficiency for HDFS 9

学校读者我要写书评

暂无评论

Energy Efficiency for HDFS

9th IEEE International Conference on Cloud computing in Emerging Markets, CCEM 2020

作者： Das, Abhishek Snv, Bhargav Sanketh Reddy, N. Phalachandra, H.L. Pes University Dept. of Computer Science and Engineering Bangalore India Center for Cloud Computing and Big Data Pes University Bangalore India

ISBN: (纸本)9781728195766

The rapidly increasing growth of big data is leading to increased usage of distributed computing and clusters as this can enable tasks to be split and processed in parallel resulting in higher computational efficiency. Environments like Apache Hadoop use Distributed file systems like the HDFS to support large clusters of typically commodity hardware to achieve the same. Commodity hardware present in HDFS consumes significant energy even when not in an active state, and considering our fragile ecosystem, reduction of the carbon footprint and conservation of energy in these environments is the focus of our work. We have evolved on a dynamic approach to Hadoop Distributed File Systems (HDFS) that provides an energy efficient paradigm for Hadoop clusters by segregating the servers in the cluster to two 'Zones', a 'Hot-Zone', and a 'Cold-Zone'. We use energy efficient storage devices (SSDs), to store data that is frequently used for the hot zone. The Cold-Zone would have lesser energy efficient devices (HDDs) and would be used to store data that is not frequently used. The Hot-Zone would always be in an active state to compute while the cold zone would have the HDDs to be in a low power consumption state. We have also enhanced the block placement policy for HDFS to support dynamic block transfer among the two zones. © 2020 IEEE.

关键词： Energy efficiency