检索结果-内蒙古大学图书馆

FedMID: A data-Free Method for Using Intermediate Outputs as a Defense Mechanism Against Poisoning Attacks in Federated Learning

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Han, Sungwon Song, Hyeonho Park, Sungwon Cha, Meeyoung School of Computing KAIST Daejeon Korea Republic of Samsung Research Seoul Korea Republic of Data Science Group Institute for Basic Science Daejeon Korea Republic of Max Planck Institute for Security and Privacy Bochum Germany

Federated learning combines local updates from clients to produce a global model, which is susceptible to poisoning attacks. Most previous defense strategies relied on vectors derived from projections of local updates on a Euclidean space;however, these methods fail to accurately represent the functionality and structure of local models, resulting in inconsistent performance. Here, we present a new paradigm to defend against poisoning attacks in federated learning using functional mappings of local models based on intermediate outputs. Experiments show that our mechanism is robust under a broad range of computing conditions and advanced attack scenarios, enabling safer collaboration among data-sensitive participants via federated learning. © 2024, CC BY.

关键词： Vector spaces

MVMoE: Multi-Task Vehicle Routing Solver with Mixture-of-Experts

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zhou, Jianan Cao, Zhiguang Wu, Yaoxin Song, Wen Ma, Yining Zhang, Jie Xu, Chi College of Computing and Data Science Nanyang Technological University Singapore School of Computing and Information Systems Singapore Management University Singapore Department of Information Systems Eindhoven University of Technology Netherlands Institute of Marine Science and Technology Shandong University China Singapore

Learning to solve vehicle routing problems (VRPs) has garnered much attention. However, most neural solvers are only structured and trained independently on a specific problem, making them less generic and practical. In this paper, we aim to develop a unified neural solver that can cope with a range of VRP variants simultaneously. Specifically, we propose a multi-task vehicle routing solver with mixture-of-experts (MVMoE), which greatly enhances the model capacity without a proportional increase in computation. We further develop a hierarchical gating mechanism for the MVMoE, delivering a good trade-off between empirical performance and computational complexity. Experimentally, our method significantly promotes zero-shot generalization performance on 10 unseen VRP variants, and showcases decent results on the few-shot setting and real-world benchmark instances. We further conduct extensive studies on the effect of MoE configurations in solving VRPs, and observe the superiority of hierarchical gating when facing out-of-distribution data. The source code is available at: https://***/RoyalSkye/Routing-MVMoE. Copyright © 2024, The Authors. All rights reserved.

关键词： Vehicle routing

Exploring the Effectiveness of Deep Learning in Audio Compression and Restoration 2

学校读者我要写书评

暂无评论

Exploring the Effectiveness of Deep Learning in Audio Compre...

2nd IEEE International Conference on Trends in Quantum computing and Emerging Business Technologies, TQCEBT 2024

作者： Pushparani, S. Rekha, K. Sashi Sivagami, V.M. Usharani, R. Jothi, M. Meenakshi College of Engineering Department of Computer Science and Engineering Tamil Nadu Chennai India Saveetha Institute of Medical and Technical Sciences Saveetha School of Engineering Department of Computer Science and Engineering Tamil Nadu Chennai India Sri Venkateswara College of Engineering Department of Information Technology SriperumbudurTamil Nadu Chennai India Srm Institute of Science and Technology Faculty of Engineering and Technology Department of Computational Intelligence School of Computing Tamil Nadu Kattankulathur India Care College of Engineering Department of Artificial Intelligence & Data Science Tamil Nadu Trichy India

ISBN: (纸本)9798350384277

Modern audio processing requires compression and restoration for streaming services and data storage and transmission. This study examines how deep learning can address these difficulties and includes an architecture flow diagram. The suggested complex design has numerous integral stages optimized for audio compression and restoration. The input layer contains raw audio data. The pre-processing module normalizes and extracts features from raw data, frequently using Mel-frequency cepstral coefficients (MFCC) to capture audio characteristics. This module improves data quality by reducing noise and filtering. The architecture's heart is the compression module, where deep learning is used. This module compresses audio data using neural networks, maybe CNNs or RNNs. This compression is quantified by compression ratio and bitrate, with hypothetical values representing audio quality trade-offs. The compression module compacts audio data for storage or transmission. To minimize data size, quantization and encoding are common here. To accurately decompress audio, the decompression module mimics the compression module's architecture. Depending on needs, error correction or quality enhancement layers may be added. Filtering, equalization, and dynamic range compression may be used in the post-processing module to improve audio quality. If needed, a restoration module uses deep learning models to minimize noise and artifacts. This study presents four hypothetical scenarios to demonstrate different results. The scenarios emphasize high compression, balanced approaches, quality prioritization, and low resource utilization. Deep learning approaches can adapt to varied audio processing needs by balancing compression ratio, bitrate, audio quality (measured by PSNR), and restoration accuracy in each case. © 2024 IEEE.

关键词： Audio streaming

Development of A Multipurpose Healthcare Mobile App Using JAVA

学校读者我要写书评

暂无评论

Development of A Multipurpose Healthcare Mobile App Using JA...

IEEE International Conference on System, Computation, Automation and Networking (ICSCAN)

作者： R K Pongiannan S. Nikkath Bushra Syed Ismail Abdul Lathif Saahil Pradhan R Brindha J Franklin Department of Computing Technologies SRM Institute of Science and Technology Kattankulathur India Data science and business systems School of computing SRM Institute of Science and Technology Kattankulathur India Department of Computational Intelligence SRM Institute of Science and Technology Kattankulathur India Department of Electrical and Electronics Engineering SRM Institute of Science and Technology Kattankulathur India

Health care has become an essential part of present human life. The preliminary requirement to select the basic information about medical facilities through a mobile app will serve the needs of the common people. This paper presents an integrated mobile app developed using JAVA. This app is user-friendly in terms of login, registration, and bookings. This caters to the facility to select information regarding doctors, medical lab facilities, pharmacy details, and health articles. This app will provide communication.

关键词：

Prompt as Free Lunch: Enhancing Diversity in Source-Free Cross-domain Few-shot Learning through Semantic-Guided Prompting

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zhuo, Linhai Wang, Zheng Qian, Tianwen Fu, Yuqian College of Computer and Data Science Fuzhou University Fuzhou China Zhejiang University of Technology College of Computer Science Zhejiang China Key Laboratory of Data Science and Intelligent Computing Hangzhou International Innovation Institute Beihang University Zhejiang China INSAIT Sofia University Bulgaria

The source-free cross-domain few-shot learning (CD-FSL) task aims to transfer pretrained models to target domains utilizing minimal samples, eliminating the need for source domain data. Addressing this issue requires models to have robust generalization abilities and strong feature representation, aligning with the characteristics of large-scale pretrained models. However, large-scale models tend to lose representational ability in cross-domain scenarios due to limited sample diversity. Given the abundant diversity provided by semantic modality, this paper utilize prompt as "free lunch" to enhance the diversity and leverages textual modality to guide the prompt generating. Specifically, we propose the SeGD-VPT framework, which is divided into two phases. The first step aims to increase feature diversity by adding diversity prompts to each support sample, thereby generating varying input and enhancing sample diversity. Furthermore, we use diversity descriptions of classes to guide semantically meaningful learning of diversity prompts, proposing random combinations and selections of texts to increase textual diversity. Additionally, deep prompt tuning is introduced to enhance the model's transfer capability. After training of the first step, support samples with different diversity prompts are input into the CLIP backbone to generate enhanced features. After generation, the second phase trains classifiers using the generated features. Extensive experimental results across several benchmarks verify our method is comparable to SOTA source-utilized models and attain the best performance under the source-free CD-FSL setting. Copyright © 2024, The Authors. All rights reserved.

关键词： Zero-shot learning

A Secured and Optimized Broadcast Authentication Scheme for the Internet of Medical Things 4

学校读者我要写书评

暂无评论

A Secured and Optimized Broadcast Authentication Scheme for ...

4th IEEE International Conference on Electrical, Computer, and Energy Technologies, ICECET 2024

作者： Riaz, Syed Waleed Khan, Fazlullah Ullah, Insaf Alshawi, Bandar Alturki, Ryan Wedyan, Mohammad Abdul Wali Khan University Mardan Computer Science Department KPK Pakistan School of Computer Science University of Nottingham Ningbo China Faculty of Science and Engineering Zhejiang Ningbo315104 China Institute for Analytics and Data Science University of Essex ColchesterCO4 3SQ United Kingdom College of Computing Umm Al-Qura University Department of Computer and Network Engineering Makkah Saudi Arabia College of Computing Umm Al-Qura University Department of Software Engineering Makkah Saudi Arabia Faculty of Information Technology and Computer Sciences Department of Computer Sciences Irbid21163 Jordan

ISBN: (纸本)9798350395914

This Internet of Medical Things (IoMT), facilitates the medical stop regarding real-time monitoring of patients, medical emergency management, remote surgery, patient information management, medical equipment, drug monitoring, etc. However, IoMT devices communicate with each other in an open environment that makes them vulnerable to a wide range of malicious threats from malicious entities. To protect the devices and their associated data, we designed a broadcast authentication scheme for IoMT devices using identity-based public key cryptography that exploits the lightweight features of the Hyper-Elliptic Curve (HEC). We then performed the security analysis based on the Random Oracle Model (ROM), in which we have proved that the proposed scheme is unforgeable under the hardiness of the hyperelliptic curve discrete logarithm problem. The proposed scheme is analyzed in terms of computational and communication overhead and the experimental results justify the superiority of the proposed work in comparison to the existing schemes. © 2024 IEEE.

关键词： Authentication

Fast Image matting based on matting pyramids and surrogate models 2

学校读者我要写书评

暂无评论

Fast Image matting based on matting pyramids and surrogate m...

2nd International Conference on Robotics, Artificial Intelligence and Intelligent Control, RAIIC 2023

作者： Liang, Yihui Gou, Hongshan Liu, Guisong Huang, Sibo University of Electronic Science and Technology of China School of Computer Science and Engineering Chengdu China University of Electronic Science and Technology of China Zhongshan Institute School of Computer Zhongshan China Guiyang Institute of Humanities and Technology College of Big Data and Information Engineering Guiyang China Southwestern University of Finance and Economics School of Computing and Artificial Intelligence Chengdu China Huizhou University Network and Information Centre Huizhou China

ISBN: (纸本)9798350328004

Image matting is a foundation task in the field of computer vision. Pixel-pair-optimization-based image matting methods have attracted lots of attention due to their distinct advantages on parallelization and handling spatially disconnected foregrounds. However, these methods struggle when the computing time is limited. To address this issue, this study presents the image matting method based on matting pyramid and surrogate model (IMMPSM). IMMPSM speeds up the process of pixel pair optimization by replaces the evolutionary algorithm in the pyramid matting framework, which consumes a lot of computing time, with the efficient surrogate-model-based method. The constructed surrogate model is reused by the presented surrogate model share strategy to save computing time. Specifically, the constructed surrogate model is shared with the pixel pair optimization subproblem adjacent to current subproblem avoiding repeated computation on the surrogate model construction. Extensive experimental results show that IMMPSM can provide acceptable alpha mattes even under limited computing time. © 2023 IEEE.

关键词： Pixels

LINKAGE: Listwise Ranking among Varied-Quality References for Non-Factoid QA Evaluation via LLMs

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Yang, Sihui Bi, Keping Cui, Wanqing Guo, Jiafeng Cheng, Xueqi CAS Key Laboratory of Network Data Science and Technology Institute of Computing Technology Chinese Academy of Sciences China University of Chinese Academy of Sciences Beijing China

Non-Factoid (NF) Question Answering (QA) is challenging to evaluate due to diverse potential answers and no objective criterion. The commonly used automatic evaluation metrics like ROUGE or BERTScore cannot accurately measure semantic similarities or answers from different perspectives. Recently, Large Language Models (LLMs) have been resorted to for NFQA evaluation due to their compelling performance on various NLP tasks. Common approaches include pointwise scoring of each candidate answer and pairwise comparisons between answers. Inspired by the evolution from pointwise to pairwise to listwise in learning-to-rank methods, we propose a novel listwise NFQA evaluation approach, that utilizes LLMs to rank candidate answers in a list of reference answers sorted by descending quality. Moreover, for NF questions that do not have multi-grade or any golden answers, we leverage LLMs to generate the reference answer list of various quality to facilitate the listwise evaluation. Extensive experimental results on three NFQA datasets, i.e., ANTIQUE, the TREC-DL-NF, and WebGLM show that our method has significantly higher correlations with human annotations compared to automatic scores and common pointwise and pairwise approaches. Our code and dataset can be found at https://***/babyyang525/ LINKAGE-Listwise-NFQA-Evaluation. Copyright © 2024, The Authors. All rights reserved.

关键词： Semantics

Computational challenges in mixed-dimensional beam/solid coupling

学校读者我要写书评

暂无评论

Computational challenges in mixed-dimensional beam/solid cou...

92nd Annual Meeting of the International Association of Applied Mathematics and Mechanics (GAMM)

作者： Firmbach, Max Steinbrecher, Ivo Popp, Alexander Mayr, Matthias Institute for Mathematics and Computer-Based Simulation University of the Bundeswehr Munich Werner-Heisenberg-Weg 39 Neubiberg85577 Germany Data Science & Computing Lab University of the Bundeswehr Munich Werner-Heisenberg-Weg 39 Neubiberg85577 Germany

The simulation of mixed-dimensional couplings poses a variety of computational challenges in order to enable accurate mod¬eling and efficient solution of problems in science, engineering and biomedicine. In this contribution, we study computational challenges and bottlenecks in the mixed-dimensional modeling of coupled beam/solid systems and discuss possible remedies to allow large-scale computations of applications in structural engineering. In particular, we address questions of memory consumption in geometric search operations, partitioning for computations on parallel clusters and performance of the linear solver by suitable preconditioning techniques. © 2023 The Authors.

关键词：