检索结果-内蒙古大学图书馆

Improving Fast Adversarial Training with Prior-Guided Knowledge

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Jia, Xiaojun Zhang, Yong Wei, Xingxing Wu, Baoyuan Ma, Ke Wang, Jue Cao, Xiaochun State Key Laboratory of Information Security Institute of Information Engineering Chinese Academy of Sciences Beijing100093 China School of Cyber Security University of Chinese Academy of Sciences Beijing100049 China AI Lab Tencent Inc. Shenzhen518000 China Institute of Artificial Intelligence Beihang University Beijing100191 China Shenzhen518172 China The School of Electronic Electrical and Communication Engineering University of Chinese Academy of Sciences Beijing100049 China School of Cyber Science and Technology Sun Yat-sen University Shenzhen Campus Shenzhen518107 China

Fast adversarial training (FAT) is an efficient method to improve robustness. However, the original FAT suffers from catastrophic overfitting, which dramatically and suddenly reduces robustness after a few training epochs. Although various FAT variants have been proposed to prevent overfitting, they require high training costs. In this paper, we investigate the relationship between adversarial example quality and catastrophic overfitting by comparing the training processes of standard adversarial training and FAT. We find that catastrophic overfitting occurs when the attack success rate of adversarial examples becomes worse. Based on this observation, we propose a positive prior-guided adversarial initialization to prevent overfitting by improving adversarial example quality without extra training costs. This initialization is generated by using high-quality adversarial perturbations from the historical training process. We provide theoretical analysis for the proposed initialization and propose a prior-guided regularization method that boosts the smoothness of the loss function. Additionally, we design a prior-guided ensemble FAT method that averages the different model weights of historical models using different decay rates. Our proposed method, called FGSM-PGK, assembles the prior-guided knowledge, i.e., the prior-guided initialization and model weights, acquired during the historical training process. Evaluations of four datasets demonstrate the superiority of the proposed method. Copyright © 2023, The Authors. All rights reserved.

关键词： Machine learning

Low Frequency Wide Bandwidth Antenna Design Using the Concept of Lumped Elements Integration 9

学校读者我要写书评

暂无评论

Low Frequency Wide Bandwidth Antenna Design Using the Concep...

9th IEEE Asia-Pacific Conference on Antennas and Propagation, APCAP 2020

作者： Xia, Yinfeng Wang, Yanyan Li, Yingsong Xue, Wei Ye, Qiubo Harbin Engineering University College of Information and Communication Engineering Harbin China Science and Technology on Complex System Control Intelligent Agent Cooperation Laboratory Beijing100074 China Jimei University School of Information Engineering Xiamen361021 China

ISBN: (纸本)9781728198057

In this paper, we will propose a super wide bandwidth low frequency antenna with a compact size for HF, VHF and UHF applications. The developed low frequency antenna consists of coplanar guide feeding structure, sleeve-like structure with lumped elements, reactance loading in the feeding. The antenna is investigated and evaluated in the HFSS, and the simulation results shows that the proposed antenna has a very wide fractional bandwidth of 191.9% from 10 MHz to 481 MHz for supporting the HF, VHF, UHF and P-band detection system applications in various platform like aircraft, ship and vehicles. © 2020 IEEE.

关键词： Bandwidth

Channel Estimation for FAS-assisted Multiuser mmWave Systems

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Xu, Hao Zhou, Gui Wong, Kai-Kit New, Wee Kiat Wang, Chao Chae, Chan-Byoung Murch, Ross Jin, Shi Zhang, Yangyang The Department of Electronic and Electrical Engineering University College London LondonWC1E7JE United Kingdom Yonsei Frontier Laboratory Yonsei University Seoul03722 Korea Republic of Erlangen91054 Germany The Integrated Service Networks Laboratory Xidian University Xi’an710071 China School of Integrated Technology Yonsei University Seoul03722 Korea Republic of Hong Kong University of Science and Technology Clear Water Bay Hong Kong The Frontiers Science Center for Mobile Information Communication and Security Southeast University Nanjing China Kuang-Chi Science Limited Hong Kong

This letter investigates the challenge of channel estimation in a multiuser millimeter-wave (mmWave) time-division duplexing (TDD) system. In this system, the base station (BS) uses a multi-antenna uniform linear array (ULA), while each mobile user has a fluid antenna system (FAS). Accurate channel state information (CSI) plays a crucial role in the precise placement of antennas in FAS. To tackle this issue, we propose a low-sample-size sparse channel reconstruction (L3SCR) method, capitalizing on the sparse propagation paths of mmWave channels. Simulation results show that our method obtains precise CSI with minimal hardware switching and pilot overhead. As a result, the system sum-rate approaches the upper bound with perfect CSI. Copyright © 2023, The Authors. All rights reserved.

关键词： Channel estimation

Weapon Target Assignment for Surface-to-air Defense with an Improved Reinforcement Genetic Algorithm

学校读者我要写书评

暂无评论

Weapon Target Assignment for Surface-to-air Defense with an ...

Chinese Automation Congress (CAC)

作者： Hua Gong Dalong Liu Yong Zhang Ke Xu School of Science Shenyang Ligong University Shenyang China School of Automation and Electrical Engineering Shenyang Ligong University Shenyang China Science and Technology on Electro-Optical Information Security Control Laboratory Tianjin China

ISBN: (纸本)9781665426480

Decision optimization in surface-to-air defense and coordinated weapon target assignment can improve the defense effect and the utilization rate of weapons. Considering the constraints of weapon capability, a multi-target optimization model is established to maximize the damage probability and minimize the weapon consumption cost. A dual population genetic algorithm improved by reinforcement learning is designed to solve this model. The chromosome exchange ratio is controlled by reinforcement learning to ensure population diversity. Experimental results verify the convergence speed and accuracy of the improved reinforcement genetic algorithm.

关键词： Costs Weapons Heuristic algorithms Sociology Reinforcement learning Statistics Optimization

Dynamic Threat Assessment of Air multi-target Based on DBN-TOPSIS Method

学校读者我要写书评

暂无评论

Dynamic Threat Assessment of Air multi-target Based on DBN-T...

Chinese Automation Congress (CAC)

作者： Hua Gong Xiaoye Yu Yong Zhang Fang Liu School of Science Shenyang Ligong University Shenyang China School of Automation and Electrical Engineering Shenyang Ligong University Shenyang China Science and Technology on Electro-Optical Information Security Control Laboratory Tianjin China

ISBN: (纸本)9781665426480

In the traditional air target threat assessment, only quantitative threat degree is considered, and the target threat range assessment is missing. Aiming at the problem, a comprehensive dynamic threat assessment method is designed in this paper. This comprehensive method considers two terms of the target threat: threat degree and threat range. In terms of threat degree assessment, a dynamic Bayesian network (DBN) and a technique for order preference by similarity to an ideal solution (TOPSIS) are fused. The multi-target fusion assessment method DBN-TOPSIS is proposed. This method maps the probability of multi-target threat level obtained by dynamic Bayesian network to TOPSIS evaluation matrix. This strategy solves the problem of multi-target threat ranking. In terms of threat range assessment, a grey prediction threat range assessment method GM(1,1) is obtained. This method solves the problem of small sample threat range assessment. The experimental results show that the multi-target fusion threat assessment method is reasonable, and the target threat range can directly reflect the change of target threat degree in airspace.

关键词： Automation Threat assessment Bayes methods

Exploring Inconsistent Knowledge Distillation for Object Detection with Data Augmentation

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Liang, Jiawei Liang, Siyuan Liu, Aishan Ma, Ke Li, Jingzhi Cao, Xiaochun School of Computer Science and Engineering Sun Yat-Sen University China State Key Laboratory of Information Security Institute of Information Engineering Chinese Academy of Sciences China NLSDE Beihang University China Institute of Dataspace Hefei China School of Electronic Electrical and Communication Engineering University of Chinese Academy of Sciences China School of Cyber Science and Technology Sun Yat-Sen University China

Knowledge Distillation (KD) for object detection aims to train a compact detector by transferring knowledge from a teacher model. Since the teacher model perceives data in a way different from humans, existing KD methods only distill knowledge that is consistent with labels annotated by human expert while neglecting knowledge that is not consistent with human perception, which results in insufficient distillation and sub-optimal performance. In this paper, we propose inconsistent knowledge distillation (IKD), which aims to distill knowledge inherent in the teacher model’s counter-intuitive perceptions. We start by considering the teacher model’s counter-intuitive perceptions of frequency and non-robust features. Unlike previous works that exploit fine-grained features or introduce additional regularizations, we extract inconsistent knowledge by providing diverse input using data augmentation. Specifically, we propose a sample-specific data augmentation to transfer the teacher model’s ability in capturing distinct frequency components and suggest an adversarial feature augmentation to extract the teacher model’s perceptions of non-robust features in the data. Extensive experiments demonstrate the effectiveness of our method which outperforms state-of-the-art KD baselines on one-stage, two-stage and anchor-free object detectors (at most +1.0 mAP). Our codes will be made available at https://***/JWLiang007/***. Copyright © 2022, The Authors. All rights reserved.

关键词： Object detection

Deterministic bulk-boundary correspondences for skin and edge modes in a general two-band non-Hermitian system

学校读者我要写书评

暂无评论

Physical Review Research 2022年第4期4卷 043222-043222页

作者： Chong Hou Lingfang Li Shihua Chen Yi Liu Luqi Yuan Yiqi Zhang Zhenhua Ni School of Physics and Frontiers Science Center for Mobile Information Communication and Security Southeast University Nanjing 211189 China Purple Mountain Laboratories Nanjing 211111 China Shanghai Key Lab of Modern Optical System University of Shanghai for Science and Technology Shanghai 200093 China CAS Center for Excellence in Ultra-Intense Laser Science Shanghai 201800 China State Key Laboratory of Advanced Optical Communication Systems and Networks School of Physics and Astronomy Shanghai Jiao Tong University Shanghai 200240 China Key Laboratory for Physical Electronics and Devices of the Ministry of Education & Shaanxi Key Lab of Information Photonic Technique School of Electronic Science and Engineering Faculty of Electronic and Information Engineering Xi'an Jiaotong University Xi'an 710049 China

In isolated Hermitian systems, the bulk-boundary correspondence is generally thought of as a compelling principle guiding the occurrence of robust edge states through the topological invariants of the bulk. However, in open non-Hermitian systems that support non-Hermitian skin effects, the universality of the bulk-boundary correspondence has so far remained disputable in spite of being subjected to intensive studies. Here we provide analytical and numerical evidence of bulk-boundary correspondences for both skin and edge modes in 1D two-band non-Hermitian systems within the framework of a prototypical non-Hermitian Su-Schrieffer-Heeger model. Two different topological winding numbers, one on a Brillouin zone and the other on a generalized Brillouin zone, are defined to fully characterize the topological phases of matter and to understand such deterministic correspondences. In addition to the exact solutions for bulk states, we also obtain explicit solution forms for edge modes along with analytical conditions for their existence and localization, all exhibiting remarkable agreement with numerical results. Our results solve the elusive non-Hermitian topology and may facilitate experimental investigations of bulk-boundary correspondence in a wide range of non-Hermitian systems.

关键词： Edge states Skin effect Topological phases of matter Non-Hermitean systems Topological materials

EMO: Edge Model Overlays to Scale Model Size in Federated Learning

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Wu, Di He, Weibo Feng, Wanglei Wen, Zhenyu Qian, Bin Varghese, Blesson School of Computer Science University of St Andrews United Kingdom Institute of Cyberspace Security College of Information Engineering Zhejiang University of Technology China University of Science and Technology of China China State Key Laboratory of Industrial Control Technology Zhejiang University China

Federated Learning (FL) trains machine learning models on edge devices with distributed data. However, the computational and memory limitations of these devices restrict the training of large models using FL. Split Federated Learning (SFL) addresses this challenge by distributing the model across the device and server, but it introduces a tightly coupled data flow, leading to computational bottlenecks and high communication costs. We propose EMO as a solution to enable the training of large models in FL while mitigating the challenges of SFL. EMO introduces Edge Model Overlay(s) between the device and server, enabling the creation of a larger ensemble model without modifying the FL workflow. The key innovation in EMO is Augmented Federated Learning (AFL), which builds an ensemble model by connecting the original (smaller) FL model with model(s) trained in the overlay(s) to facilitate horizontal or vertical scaling. This is accomplished through three key modules: a hierarchical activation replay cache to decouple AFL from FL, a convergence-aware communication controller to optimize communication overhead, and an ensemble inference module. Evaluations on a real-world prototype show that EMO improves accuracy by up to 17.77% compared to FL, and reduces communication costs by up to 7.17× and decreases training time by up to 6.9× compared to SFL. © 2025, CC BY.

关键词： Federated learning

Ground-to-air-Defense Firepower Assignment Decision Based on Non-cooperative Game

学校读者我要写书评

暂无评论

Ground-to-air-Defense Firepower Assignment Decision Based on...

Chinese Automation Congress (CAC)

作者： Hua Gong Mengya Cui Yong Zhang Wenjuan Sun School of Science Shenyang Ligong University Shenyang China School of Automation and Electrical Engineering Shenyang Ligong University Shenyang China Science and Technology on Electro-Optical Information Security Control Laboratory Tianjin China

ISBN: (纸本)9781665426480

The key problem of ground-to-air defense is how to make accurate and effective firepower assignment decision for attacking targets. A multi-objective optimization model of ground-to-air defense firepower assignment is established with the aim of maximizing the damage probability and minimizing the weapon consumption, and non-cooperative game theory is used to solve the problem. A non-cooperative game model is established by mapping the two objectives to players, and weapon assignment schemes to strategy profiles. An improved tabu genetic algorithm based on Nash equilibrium is designed to get the assignment scheme. Experimental results show that the proposed method has better convergence and stability than the improved genetic algorithm based on tabu search for multi-objective optimization model.

关键词： Automation Weapons Games Nash equilibrium Search problems Minimization Stability analysis