检索结果-内蒙古大学图书馆

Probing the Robustness of Vision-Language Pretrained Models: A Multimodal Adversarial Attack Approach

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Guan, Jiwei Ding, Tianyu Cao, Longbing Pan, Lei Wang, Chen Zheng, Xi School of Computing Macquarie University Sydney Australia Applied Sciences Group Microsoft Corporation Redmond United States School of Information Technology Deakin University Waurn Ponds Australia Data61 CSIRO Sydney Australia

Vision-language pretraining (VLP) with transformers has demonstrated exceptional performance across numerous multimodal tasks. However, the adversarial robustness of these models has not been thoroughly investigated. Existing multimodal attack methods have largely overlooked cross-modal interactions between visual and textual modalities, particularly in the context of cross-attention mechanisms. In this paper, we study the adversarial vulnerability of recent VLP transformers and design a novel Joint Multimodal Transformer Feature Attack (JMTFA) that concurrently introduces adversarial perturbations in both visual and textual modalities under white-box settings. JMTFA strategically targets attention relevance scores to disrupt important features within each modality, generating adversarial samples by fusing perturbations and leading to erroneous model predictions. Experimental results indicate that the proposed approach achieves high attack success rates on vision-language understanding and reasoning downstream tasks compared to existing baselines. Notably, our findings reveal that the textual modality significantly influences the complex fusion processes within VLP transformers. Moreover, we observe no apparent relationship between model size and adversarial robustness under our proposed attacks. These insights emphasize a new dimension of adversarial robustness and underscore potential risks in the reliable deployment of multimodal AI systems. Copyright © 2024, The Authors. All rights reserved.

关键词： Generative adversarial networks

Priority Based Fair Scheduling: Enhancing Efficiency in Cloud Job Distribution

学校读者我要写书评

暂无评论

Priority Based Fair Scheduling: Enhancing Efficiency in Clou...

International Conference on Software Engineering and Computer Systems (ICSECS)

作者： Saydul Akbar Murad Zafril Rizal M Azmi Faria Jerin Brishti Md Saib Anupam Kumar Bairagi Faculty of Computing College of Computing & Applied Sciences Universiti Malaysia Pahang Al-Sultan Abdullah Pahang Malaysia Computer Science and Engineering International Islamic University Chittagong Chattogram Bangladesh South China University of Technology Guangzhou China Khulna University Khulna Bangladesh

In recent years, there has been a growing interest in cloud computing as a means to enhance user access to shared computing resources, including software and hardware, through the internet. However, the efficient utilization of these cloud resources has been a challenge, often resulting in wastage or degraded service performance due to inadequate scheduling. To overcome this challenge, numerous researchers have focused on improving existing Priority Rule (PR) cloud schedulers by developing dynamic scheduling algorithms, but they have fallen short of meeting user satisfaction. In this study, we introduce a new PR scheduler called Priority Based Fair Scheduling (PBFS), which takes into account key parameters such as CPU Time, Job Arrival Time, and Job Length. We evaluate the performance of PBFS by comparing it with five existing algorithms, and the results demonstrate that PBFS surpasses the performance of the other algorithms. The experiment was conducted using the CloudSim simulator, utilizing a dataset of 300 and 400 jobs. In order to assess the performance, three key metrics were employed: flow time, makespan time, and total tardiness. These metrics were chosen to evaluate and analyze the effectiveness of the proposed scheduling algorithm.

关键词：

A Comparison of Low Complexity CNN Models and Data Enhancement for Efficient Android Implementation of Leaves Recognition

学校读者我要写书评

暂无评论

A Comparison of Low Complexity CNN Models and Data Enhanceme...

International Symposium on Advanced Topics in Electrical Engineering (ATEE)

作者： Alin-Gabriel Cococi Radu Dogaru Doctoral School in Electronics Telecommunications and Information Technology University "Politehnica" of Bucharest Romania Dept. of Applied Electronics and Information Eng Natural Computing Laboratory University "Politehnica" of Bucharest Romania

Classification of flora data is a high complexity problem due to the similarity of plants. Identification and determination of plant species is a task that most of the general public is unable to do and it can be a challenge even for the expertise of qualified individuals. This paper studies the possibility of using lightweight convolutional neural networks on resources constrained devices or platforms to accurately classify leaves of plants. Several low complexity convolutional neural networks are considered for the problem of leaves recognition in terms of accuracy, precision, latency and complexity. A novel scheme for enhancement and augmentation of the leaf’s images is proposed and demonstrated as capable to improve performance. MobileNet, L-CNN and NL-CNN models with Android implementations in Tensorflow Lite1 are considered, demonstrating the capability to build a portable intelligent instrument capable to identify plants by their leaves with an accuracy of 95.3%.

关键词： Performance evaluation Electrical engineering Electric potential Databases Instruments Data models Complexity theory

MIXED-INTEGER OPTIMISATION OF GRAPH NEURAL NETWORKS FOR COMPUTER-AIDED MOLECULAR DESIGN

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Schweidtmann, Artur M. McDonald, Tom Tsay, Calvin Yorke-Smith, Neil Delft Institute of Applied Mathematics Delft University of Technology Delft Netherlands Department of Computing Imperial College London London United Kingdom Department of Chemical Engineering Delft University of Technology Delft Netherlands STAR Lab Delft University of Technology Delft Netherlands

ReLU neural networks have been modelled as constraints in mixed integer linear programming (MILP), enabling surrogate-based optimisation in various domains and efficient solution of machine learning certification problems. However, previous works are mostly limited to MLPs. Graph neural networks (GNNs) can learn from non-euclidean data structures such as molecular structures efficiently and are thus highly relevant to computer-aided molecular design (CAMD). We propose a bilinear formulation for ReLU Graph Convolutional Neural Networks and a MILP formulation for ReLU GraphSAGE models. These formulations enable solving optimisation problems with trained GNNs embedded to global optimality. We apply our optimization approach to an illustrative CAMD case study where the formulations of the trained GNNs are used to design molecules with optimal boiling *** Codes 90C11 © 2023, CC BY-NC-SA.

关键词： Integer programming

Classification of Retroviruses Based on Genomic Data Using RVGC

学校读者我要写书评

暂无评论

Computers, Materials & Continua 2021年第12期69卷 3829-3844页

作者： Khalid Mahmood Aamir Muhammad Bilal Muhammad Ramzan Muhammad Attique Khan Yunyoung Nam Seifedine Kadry Department of CS&IT University of SargodhaSargodha40100Pakistan Department of CS&IT University of MianwaliMianwali42200Pakistan School of Systems and Technology University of Management and TechnologyLahore54782Pakistan Department of Computer Science HITEC University TaxilaTaxilaPakistan Department of Computer Science and Engineering Soonchunhyang UniversityAsanKorea Faculty of Applied Computing and Technology Noroff University CollegeKristiansandNorway

Retroviruses are a large group of infectious agents with similar virion structures and replication ***,cancer,neurologic disorders,and other clinical conditions can all be fatal due to retrovirus *** of retroviruses by genome sequence is a biological problem that benefits from computational *** National Center for Biotechnology Information(NCBI)promotes science and health by making biomedical and genomic data available to the *** research aims to classify the different types of rotavirus genome sequences available at the ***,nucleotide pattern occurrences are counted in the given genome sequences at the preprocessing *** on some significant results,the number of features used for classification is reduced to *** classification shall be carried out in two *** first phase of classification shall select only two *** data in the first phase is transferred to the next phase,where the final decision is taken with the remaining three *** data sets of animals and human retroviruses are selected;the training data set is used to minimize the classifier’s number and training;the validation data set is used to validate the *** performance of the classifier is analyzed using the test data ***,we use decision tree,naive Bayes,knearest neighbors,and vector support machines to compare *** results show that the proposed approach performs better than the existing methods for the retrovirus’s imbalanced genome-sequence dataset.

关键词： Retroviruses machine learning bioinformatics classification

Implicit U-KAN2.0: Dynamic, Efficient and Interpretable Medical Image Segmentation

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Cheng, Chun-Wun Zhao, Yining Cheng, Yanqi Montoya, Javier Schönlieb, Carola-Bibiane Aviles-Rivero, Angelica I. Department of Applied Mathematics and Theoretical Physics University of Cambridge United Kingdom Siebel School of Computing and Data Science University of Illinois Urbana-Champaign United States Lucerne School of Computer Science and Information Technology Lucerne University of Applied Sciences and Arts United States Yau Mathematical Sciences Center Tsinghua University China

Image segmentation is a fundamental task in both image analysis and medical applications. State-of-the-art methods predominantly rely on encoder-decoder architectures with a U-shaped design, commonly referred to as U-Net. Recent advancements integrating transformers and MLPs improve performance but still face key limitations, such as poor interpretability, difficulty handling intrinsic noise, and constrained expressiveness due to discrete layer structures, often lacking a solid theoretical *** this work, we introduce Implicit U-KAN 2.0, a novel U-Net variant that adopts a two-phase encoder-decoder structure. In the SONO phase, we use a second-order neural ordinary differential equation (NODEs), called the SONO block, for a more efficient, expressive, and theoretically grounded modeling approach. In the SONO-MultiKAN phase, we integrate the second-order NODEs and MultiKAN layer as the core computational block to enhance interpretability and representation power. Our contributions are threefold. First, U-KAN 2.0 is an implicit deep neural network incorporating MultiKAN and second order NODEs, improving interpretability and performance while reducing computational costs. Second, we provide a theoretical analysis demonstrating that the approximation ability of the MultiKAN block is independent of the input dimension. Third, we conduct extensive experiments on a variety of 2D and a single 3D dataset, demonstrating that our model consistently outperforms existing segmentation networks. Copyright © 2025, The Authors. All rights reserved.

关键词： Deep neural networks

Multi-Layer Perceptron Regression Model for Material Characterization in the Microwave Frequency Range with Cross-Dataset Validation 7

学校读者我要写书评

暂无评论

Multi-Layer Perceptron Regression Model for Material Charact...

7th International Youth Conference on Radio Electronics, Electrical and Power Engineering, REEPE 2025

作者： Hanima Kannan, C.H. Gupta, Deepa Mathur, Parul Kurup, Dhanesh G. Chaudhari, Tilesh Sanjay Mathur, Priyanka Amrita School of Engineering Amrita Vishwa Vidyapeetham Rf and Wireless Systems Laboratory Department of Electronics and Communication Engineering Bengaluru India Amrita School of Computing Amrita Vishwa Vidyapeetham Department of Computer Science and Engineering Bengaluru India Indian Institute of Technology Creative and Advanced Research Based on Nanomaterials Laboratory Department of Chemical Engineering Hyderabad India School of Engineering and Applied Science Bennett University U.P. Greater Noida India

ISBN: (纸本)9798331531836

This paper presents a Multi-Layer Perceptron (MLP) based regression model for characterizing materials at micrwave frequency range. A mathematical model and experimental setup of the Open-Ended Coaxial Probe (OECP) sensor interacting with the material under test (MUT) generate a cross-dataset to validate the MLP regression model, utilizing the frequency-dependent complex dielectric constant for material characterization. This paper utilizes the frequency-dependent response of the OECP sensor (scattering parameters) as input features for the MLP model, while the complex dielectric constant of the MUT serves as the output feature. The optimized MLP model has an R2 of 99.991% and MSE of 6.65 × 10-5, with predictions closely aligning with literature values for standard loads such as distilled water, acetone, and ethanol, obtained using a commercially available dielectric assessment software kit. This cost-effective model enables precise dielectric property estimation, with future work extending to unknown materials, mixtures, and electrolyte solutions. © 2025 IEEE.

关键词： Linear regression

Fighting typhoid fever: Modeling antibiotic resistance and antibiotic switching

学校读者我要写书评

暂无评论

Franklin Open

Franklin Open 2025年 11卷

作者： Shaale, Maria M. Mushanyu, Josiah Nyabadza, Farai Nuugulu, Samuel M. Department of Computing Mathematical & Statistical Science University of Namibia Windhoek 13301 Namibia Department of Mathematics and Applied Mathematics University of Johannesburg Johannesburg 2006 South Africa Institute of Applied Research and Technology Emirates Aviation University Dubai International Academic City United Arab Emirates

Typhoid fever continues to be a major public health concern, particularly in developing countries where the sanitation infrastructure is inadequate. The rise in resistance to typhoid drugs has made treatment increasingly challenging, resulting in longer recovery times and continued transmission of the disease within households and communities. This growing resistance underscores the urgent need for improved treatment strategies and public health intervention. In this study, we presented a mathematical model of typhoid fever that incorporates antibiotic resistance and the implementation of antibiotic switching as a control strategy. The model considers individuals infected with typhoid antibiotic sensitive strains and typhoid antibiotic resistant strain. The effects of antibiotic switching, which involves transitioning patients between different antibiotics, are modeled to study its impact on the prevalence of resistant and sensitive strains. The model is analyzed and the model reproduction number, R₀, is found to be the sum of two reproduction numbers R_s and R_r representing the contribution of the sensitive and resistant strains, respectively. The stability analysis indicates that the disease-free equilibrium is stable when the model reproduction number is less than one, suggesting the possibility of eradicating the disease under effective control measures. In contrast, the endemic equilibrium remains stable when the reproduction number exceeds one, indicating persistent infection levels. Sensitivity analysis is performed to identify critical parameters that influence the persistence of typhoid in the population. Numerical simulations are performed to support the theoretical findings. The results obtained demonstrate that antibiotic switching can reduce the prevalence of resistant and sensitive strains and overall infection levels, highlighting their potential as an effective strategy to manage antibiotic resistance in typhoid fever.

关键词： Antibiotic resistance Numerical simulations Reproduction number Strains Switching Typhoid

UNRAVELING THE MOLECULAR MAGIC: AI INSIGHTS ON THE FORMATION OF EXTRAORDINARILY STRETCHABLE HYDROGELS

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Emmami, Shahriar Hojjati Meibody, Ali Pilehvar Tayebi, Lobat Tavakoli, Mohammadamin Baldi, Pierre Department of Biomedical Engineering Amirkabir University of Technology Iran Department of Applied Science and Technology Polytechnic University of Turin Italy Department of Developmental Sciences Marquette University United States Department of Computing and Mathematical Sciences California Institute of Technology United States Department of Computer Science University of California Irvine United States

The deliberate manipulation of ammonium persulfate, methylenebisacrylamide, dimethyleacrylamide, and polyethylene oxide concentrations resulted in the development of a hydrogel with an exceptional stretchability, capable of extending up to 260 times its original length. This study aims to elucidate the molecular architecture underlying this unique phenomenon by exploring potential reaction mechanisms, facilitated by an artificial intelligence prediction system. Artificial intelligence predictor introduces a novel approach to interlinking two polymers, involving the formation of networks interconnected with linear chains following random chain scission. This novel configuration leads to the emergence of a distinct type of hydrogel, herein referred to as a "Span Network." Additionally, Fourier-transform infrared spectroscopy (FTIR) is used to investigate functional groups that may be implicated in the proposed mechanism, with ester formation confirmed among numerous hydroxyl end groups obtained from chain scission of PEO and carboxyl groups formed on hydrogel networks. © 2024, CC BY.

关键词： Hydrogels