检索结果-南通市图书馆

Mitigate noisy data for smart IoT via GAN based machine unlearning

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2024年第3期67卷 62-78页

作者： Zhuo MA Yilong YANG Yang LIU Xinjing LIU Jianfeng MA School of Cyber Engineering Xidian University state key laboratory of integrated services networks (ISN)

With the development of IoT applications, machine learning dramatically improves the utility of variable IoT systems such as autonomous driving. Although the pretrain-finetune framework can cope well with data heterogeneity in complex IoT scenarios, the data collected by sensors often contain unexpected noisy data, e.g., out-of-distribution(OOD) data, which leads to the reduced performance of fine-tuned models. To resolve the problem, this paper proposes Mu GAN, a method that can mitigate the side-effect of OOD data via the generative adversarial network(GAN)-based machine unlearning. Mu GAN follows a straightforward but effective idea to mitigate the performance loss caused by OOD data, i.e., “flashbacking” the model to the condition where OOD data are excluded from model training. To achieve the goal, we design an adversarial game, where a discriminator is trained to identify whether a sample belongs to the training set by observing the confidence score. Meanwhile, a generator(i.e., the target model) is updated to fool the discriminator into believing that the OOD data are not included in the training set but others do. The experimental results show that benefiting from the high unlearning rate(more than 90%) and retention rate(99%), Mu GAN succeeds in lowering the model performance degradation caused by OOD data from 5.88% to 0.8%.

关键词： machine unlearning generative adversarial network out of distribution data Internet of Thing neural network

Conversion of a single-layer ANN to photonic SNN for pattern recognition

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2024年第1期67卷 261-270页

作者： Yanan HAN Shuiying XIANG Tianrui ZHANG Yahui ZHANG Xingxing GUO Yuechun SHI state key laboratory of integrated Service networks State Key Discipline Laboratory of Wide Bandgap Semiconductor TechnologyXidian University Yongjiang laboratory

This work presents a complete conversion scheme for photonic spiking neural networks(SNNs).We verified that the output of an artificial neural network(ANN) trained with the simulated optical activation function can be directly converted into the spike rate of a photonic spiking neuron model.To reveal the feasibility of hardware implementation,we considered the effects of different bit precisions of data and weight,noise level,and bias current mismatch on the converted results.The proposed scheme was evaluated using the Deterding vowel,IRIS,TIDIGITS,and MNIST datasets for pattern recognition,and achieved mean accuracies of 95.80%,98.67%,96.19%,and 92.33%,respectively.The proposed scheme can convert an ANN into a photonic SNN with almost no precision loss,and the performance was comparable to that of an ANN trained with the rectified linear unit function.The proposed scheme can enable the high-performance implementation of photonic SNNs.

关键词： photonic SNN conversion optical computing pattern recognition artificial neural network

Tiered clustering-based management architecture in mega-satellite networks

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2024年第5期67卷 215-227页

作者： Qi HAO Di ZHOU Min SHENG Yan SHI Jiandong LI state key laboratory of integrated Service networks Xidian University

Hybrid orbital networking and the proliferation of satellites raise explosive management complexity in mega-satellite networks(MSNs) and prompt the management architecture to the hierarchy. In this regard, we propose a tiered clustering-based management architecture(TCMA) to realize the high-efficient and low-cost management performance. Specifically, based on the constrained satellite coverage, we derive an expression of the mapping relationship between the required control overhead of the controller interaction,the average path length of the proposed TCMA, and the corresponding number of management tiers and cluster size. The expression can determine the optimal number of tiers of the TCMA by minimizing the required flow table size between controllers, while ensuring high transmission efficiency. We then exhibit the elasticity of the hierarchical TCMA by adding the peer connections at the second-highest management tier to reduce maximum redundant paths. Simulation results verify the effectiveness of the proposed TCMA and illustrate that a 3-tier TCMA is sufficient for thousands of satellites.

关键词： mega-satellite networks hierarchical architecture average hop tiered structure clustering

Resource Allocation in Multi-User Cellular networks:A Transformer-Based Deep Reinforcement Learning Approach

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

China Communications 2024年第5期21卷 77-96页

作者： Zhao Di Zheng Zhong Qin Pengfei Qin Hao Song Bin state key laboratory of integrated services networks Xidian University710071China China Academy of Space Technology Beijing 100081China

To meet the communication services with diverse requirements,dynamic resource allocation has shown increasing importance.In this paper,we consider the multi-slot and multi-user resource allocation(MSMU-RA)in a downlink cellular scenario with the aim of maximizing system spectral efficiency while guaranteeing user fairness.We first model the MSMURA problem as a dual-sequence decision-making process,and then solve it by a novel Transformerbased deep reinforcement learning(TDRL)approach.Specifically,the proposed TDRL approach can be achieved based on two aspects:1)To adapt to the dynamic wireless environment,the proximal policy optimization(PPO)algorithm is used to optimize the multi-slot RA strategy.2)To avoid co-channel interference,the Transformer-based PPO algorithm is presented to obtain the optimal multi-user RA scheme by exploring the mapping between user sequence and resource sequence.Experimental results show that:i)the proposed approach outperforms both the traditional and DRL methods in spectral efficiency and user fairness,ii)the proposed algorithm is superior to DRL approaches in terms of convergence speed and generalization performance.

关键词： dynamic resource allocation multi-user cellular network spectrum efficiency user fairness

维普期刊数据库

Experimental and numerical demonstration of hierarchical time-delay reservoir computing based on cascaded VCSELs with feedback and multiple injections

在线全文

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2024年第2期67卷 279-290页

作者： Xingxing GUO Shuiying XIANG Xingyu CAO Biling GU state key laboratory of integrated Service networks Xidian University state key Discipline laboratory of Wide Bandgap Semiconductor Technology School of MicroelectronicsXidian University

In this paper, we propose and demonstrate experimentally and numerically a hierarchical timedelay optical reservoir computing(RC) system based on cascaded vertical-cavity surface-emitting lasers(VCSELs) with feedback and multiple injections. The prediction performance characteristics of the hierarchical time-delay RC system based on cascaded VCSELs under different reservoir layers are compared.Evidently, the prediction performance of the hierarchical time-delay RC system is first improved and saturates as the number of reservoir layers increases. Besides, the impacts of key factors on predicting the hierarchical time-delay RC system are also analyzed in detail experimentally and numerically. This proposed hierarchical time-delay RC system based on VCSELs is useful for the further development of RC systems and may be beneficial to improve the ability of RC systems to solve more complex problems.

关键词： vertical-cavity surface-emitting laser reservoir computing neuromorphic photonics time-delay system information processing

Adaptive Bistable Stochastic Resonance Based Weak Signal Reception in Additive Laplacian Noise

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

China Communications 2024年第1期21卷 228-241页

作者： Jin Liu Zan Li Qiguang Miao Li Yang School of Computer Science and Technology Xidian UniversityXi’an 710071China state key laboratory of integrated services networks Xidian UniversityXi’an 710071China

Weak signal reception is a very important and challenging problem for communication systems especially in the presence of non-Gaussian noise,and in which case the performance of optimal linear correlated receiver degrades dramatically.Aiming at this,a novel uncorrelated reception scheme based on adaptive bistable stochastic resonance(ABSR)for a weak signal in additive Laplacian noise is investigated.By analyzing the key issue that the quantitative cooperative resonance matching relationship between the characteristics of the noisy signal and the nonlinear bistable system,an analytical expression of the bistable system parameters is derived.On this basis,by means of bistable system parameters self-adaptive adjustment,the counterintuitive stochastic resonance(SR)phenomenon can be easily generated at which the random noise is changed into a benefit to assist signal transmission.Finally,it is demonstrated that approximately 8dB bit error ratio(BER)performance improvement for the ABSR-based uncorrelated receiver when compared with the traditional uncorrelated receiver at low signal to noise ratio(SNR)conditions varying from-30dB to-5dB.

关键词： adaptive bistable stochastic resonance additive Laplacian noise low signal to noise ratio uncorrelated reception scheme weak signal reception

维普期刊数据库

Experimental demonstration of a photonic spiking neuron based on a DFB laser subject to side-mode optical pulse injection

在线全文

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2024年第3期67卷 271-280页

作者： Shuiying XIANG Shuang GAO Yuechun SHI Yuna ZHANG Ziwei SONG Xingxing GUO Yahui ZHANG Yuxin MA Xiangfei CHEN state key laboratory of integrated Service networks Xidian University Yongjiang laboratory 3. College of Engineering and Applied Sciences Nanjing University

We proposed and experimentally demonstrated a simple and novel photonic spiking neuron based on a distributed feedback(DFB) laser subject to side-mode optical pulse injection(SMOPI). The DFB laser chip is designed and fabricated based on asymmetric equivalent π phase shift(π-EPS) with the reconstruction-equivalent-chirp(REC) technique. Under side-mode continuous-wave(CW) optical injection,excitability pulse was experimentally observed during the dominant mode switching process due to the injection-locked effect. Based on the transition between the excitability regime and the side-mode injection locking effect, the controllable and repeatable neuron-like spiking response can be realized when external stimulus pulses are electro-optically modulated on the CW optical carrier. The experimental results show that the spike threshold, temporal integration, and refractory period, which are important spike processing mechanisms in biological neurons, can all be achieved in the optically-injected DFB laser. The experimental findings are also verified numerically with a rate equation model that considers the SMOPI. To the best of our knowledge, this is the first experimental demonstration of a photonic spiking neuron based on a DFB laser subject to SMOPI, which holds promise for realizing large-scale photonic spiking neuron arrays for hardware photonic spiking neural network chips.

关键词： neuromorphic photonics photonic spiking neuron DFB laser side-mode optical pulse injections

Optimizing Optical Fiber Faults Detection:A Comparative Analysis of Advanced Machine Learning Approaches

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Computers, Materials & Continua 2024年第5期79卷 2697-2721页

作者： Kamlesh Kumar Soothar Yuanxiang Chen Arif Hussain Magsi Cong Hu Hussain Shah School of Electronics Engineering Beijing University of Posts and TelecommunicationsBeijing100876China state key laboratory of integrated services networks Xidian UniversityXi’an710071China state key laboratory ofNetworking and Switching Technology Beijing University of Posts and TelecommunicationsBeijing100876China

Efficient optical network management poses significant importance in backhaul and access network communicationfor preventing service disruptions and ensuring Quality of Service(QoS)satisfaction.The emerging faultsin optical networks introduce challenges that can jeopardize the network with a variety of faults.The existingliterature witnessed various partial or inadequate solutions.On the other hand,Machine Learning(ML)hasrevolutionized as a promising technique for fault detection and prevention.Unlike traditional fault managementsystems,this research has three-fold contributions.First,this research leverages the ML and Deep Learning(DL)multi-classification system and evaluates their accuracy in detecting six distinct fault types,including fiber cut,fibereavesdropping,splicing,bad connector,bending,and PC connector.Secondly,this paper assesses the classificationdelay of each classification algorithm.Finally,this work proposes a fiber optics fault prevention algorithm thatdetermines to mitigate the faults accordingly.This work utilized a publicly available fiber optics dataset namedOTDR_Data and applied different ML classifiers,such as Gaussian Naive Bayes(GNB),Logistic Regression(LR),Support Vector Machine(SVM),K-Nearest Neighbor(KNN),Random Forest(RF),and Decision Tree(DT).Moreover,Ensemble Learning(EL)techniques are applied to evaluate the accuracy of various classifiers.In addition,this work evaluated the performance of DL-based Convolutional Neural Network and Long-Short Term Memory(CNN-LSTM)hybrid classifier.The findings reveal that the CNN-LSTM hybrid technique achieved the highestaccuracy of 99%with a delay of 360 s.On the other hand,EL techniques improved the accuracy in detecting fiberoptic faults.Thus,this research comprehensively assesses accuracy and delay metrics for various classifiers andproposes the most efficient attack detection system in fiber optics.

关键词： Fiber optics fault detection multiclassification machine learning ensemble learning

Pyramid-resolution person restoration for cross-resolution person re-identification

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2024年第6期67卷 451-452页

作者： Chunlei PENG Bo WANG Decheng LIU Nannan WANG Xinbo GAO state key laboratory of integrated services networks Xidian University School of Cyber Engineering Xidian University School of Telecommunications Engineering Xidian University Chongqing key laboratory of Image Cognition Chongqing University of Posts and Telecommunications

Person images captured by surveillance cameras are typically with various resolutions and are susceptible to pose changes,occlusion and background variations,thus affecting the performance of person Re ID.Existing approaches[1,2]either use an image restoration model to restore the lowresolution images to obtain a corresponding high-resolution image,

关键词： Restoration