Hybrid precoder design is a key technique providing better antenna gain and reduced hardware complexity in millimeter-wave(mmWave)massive multiple-input multiple-output(MIMO)systems.In this paper,Gaussian Mixture lear...Hybrid precoder design is a key technique providing better antenna gain and reduced hardware complexity in millimeter-wave(mmWave)massive multiple-input multiple-output(MIMO)systems.In this paper,Gaussian Mixture learned approximate message passing(GM-LAMP)network is presented for the design of optimal hybrid precoders suitable for mmWave Massive MIMO systems.Optimal hybrid precoder designs using a compressive sensing scheme such as orthogonal matching pursuit(OMP)and its derivatives results in high computational complexity when the dimensionality of the sparse signal is high.This drawback can be addressed using classical iterative algorithms such as approximate message passing(AMP),which has comparatively low computational complexity.The drawbacks of AMP algorithm are fixed shrinkage parameter and non-consideration of prior distribution of the hybrid precoders.In this paper,the fixed shrinkage parameter problem of the AMP algorithm is addressed using learned AMP(LAMP)network,and is further enhanced as GMLAMP network using the concept of Gaussian Mixture distribution of the hybrid precoders.The simula-tion results show that the proposed GM-LAMP network achieves optimal hybrid precoder design with enhanced achievable rates,better accuracy and low computational complexity compared to the existing algorithms.展开更多
This study investigates the impact of the invariance of feature vectors for partial-to-partial point set registration under translation and rotation of input point sets,particularly in the realm of techniques based on...This study investigates the impact of the invariance of feature vectors for partial-to-partial point set registration under translation and rotation of input point sets,particularly in the realm of techniques based on deep learning and Gaussian mixture models(GMMs).We reveal both theoretical and practical problems associated with such deeplearning-based registration methods using GMMs,with a particular focus on the limitations of DeepGMR,a pioneering study in this line,to the partial-topartial point set registration.Our primary goal is to uncover the causes behind such methods and propose a comprehensible solution for that.To address this,we introduce an attention-based reference point shifting(ARPS)layer,which robustly identifies a common reference point of two partial point sets,thereby acquiring transformation-invariant features.The ARPS layer employs a well-studied attention module to find a common reference point rather than the overlap region.Owing to this,it significantly enhances the performance of DeepGMR and its recent variant,UGMMReg.Furthermore,these extension models outperform even prior deep learning methods using attention blocks and Transformer to extract the overlap region or common reference points.We believe these findings provide deeper insights into registration methods using deep learning and GMMs.Our source code and datasets are available at http://gffzzd16a119ca6e245c5su6kcq0q6v6w660un.ffgz.tsg.suse.edu.cn/DGRM-ARPS.git.展开更多
在伪造音频检测任务中,高斯混合模型(GMM)虽然可以估计每个音频帧的概率分数,但未能充分利用高斯分量的概率分布信息,影响了检测性能。为此,本研究提出了一种基于联合特征建模与创新网络架构的双分支伪造音频检测方法。首先,结合线性频...在伪造音频检测任务中,高斯混合模型(GMM)虽然可以估计每个音频帧的概率分数,但未能充分利用高斯分量的概率分布信息,影响了检测性能。为此,本研究提出了一种基于联合特征建模与创新网络架构的双分支伪造音频检测方法。首先,结合线性频率倒谱系数(LFCC)和逆梅尔频率倒谱系数(IMFCC),设计了两种基于高斯概率分数的特征:L-GPF(LFCC-based Gaussian Probability Feature)和I-GPF(IMFCCbased Gaussian Probability Feature),分别捕捉音频的低频和高频特征。其次,提出了一种叠式双向长短时记忆网络(SBi-LSTM),分别对两类特征进行建模,从帧序列中学习真实与伪造音频的特征分布差异。最后,设计了创新的联合模块,通过动态权重融合和阈值因子判别机制整合双分支网络的特征信号,显著提高分类性能。实验在ASVspoof 2019数据集上进行。针对逻辑访问场景的开发集实验结果表明,相比于只使用I-GPF作为前端特征的单一SBi-LSTM模型,双分支模型的等错误率(EER)可降低94.6%,级联检测代价函数(t-DCF)可降低76.9%。相比于传统联合方式,在物理访问场景下联合模块最高可实现EER降低33.7%,t-DCF降低29.9%。相比于LFCC+GMM的基准模型,双分支模型的等错误率降低89.9%,级联检测代价函数降低89.2%。对比实验结果表明,与最新模型相比,双分支模型的EER降低15.6%,t-DCF降低29.8%。展开更多
Smart Agriculture,also known as Agricultural 5.0,is expected to be an integral part of our human lives to reduce the cost of agricultural inputs,increasing productivity and improving the quality of the final product.I...Smart Agriculture,also known as Agricultural 5.0,is expected to be an integral part of our human lives to reduce the cost of agricultural inputs,increasing productivity and improving the quality of the final product.Indeed,the safety and ongoing maintenance of Smart Agriculture from cyber-attacks are vitally important.To provide more comprehensive protection against potential cyber-attacks,this paper proposes a new deep learning-based intrusion detection system for securing Smart Agriculture.The proposed Intrusion Detection System IDS,namely GMLPIDS,combines the feedforward neural network Multilayer Perceptron(MLP)and the Gaussian Mixture Model(GMM)that can better protect the Smart Agriculture system.GMLP-IDS is evaluated with the CIC-DDoS2019 dataset,which contains various Distributed Denial-of-Service(DDoS)attacks.The paper first uses the Pearson’s correlation coefficient approach to determine the correlation between the CIC-DDoS2019 dataset characteristics and their corresponding class labels.Then,the CIC-DDoS2019 dataset is divided randomly into two parts,i.e.,training and testing.75%of the data is used for training,and 25%is employed for testing.The performance of the newly proposed IDS has been compared to the traditional MLP model in terms of accuracy rating,loss rating,recall,and F1 score.Comparisons are handled on both binary and multi-class classification problems.The results revealed that the proposed GMLP-IDS system achieved more than 99.99%detection accuracy and a loss of 0.02%compared to traditional MLP.Furthermore,evaluation performance demonstrates that the proposed approach covers a more comprehensive range of security properties for Smart Agriculture and can be a promising solution for detecting unknown DDoS attacks.展开更多
摘要Hybrid precoder design is a key technique providing better antenna gain and reduced hardware complexity in millimeter-wave(mmWave)massive multiple-input multiple-output(MIMO)systems.In this paper,Gaussian Mixture learned approximate message passing(GM-LAMP)network is presented for the design of optimal hybrid precoders suitable for mmWave Massive MIMO systems.Optimal hybrid precoder designs using a compressive sensing scheme such as orthogonal matching pursuit(OMP)and its derivatives results in high computational complexity when the dimensionality of the sparse signal is high.This drawback can be addressed using classical iterative algorithms such as approximate message passing(AMP),which has comparatively low computational complexity.The drawbacks of AMP algorithm are fixed shrinkage parameter and non-consideration of prior distribution of the hybrid precoders.In this paper,the fixed shrinkage parameter problem of the AMP algorithm is addressed using learned AMP(LAMP)network,and is further enhanced as GMLAMP network using the concept of Gaussian Mixture distribution of the hybrid precoders.The simula-tion results show that the proposed GM-LAMP network achieves optimal hybrid precoder design with enhanced achievable rates,better accuracy and low computational complexity compared to the existing algorithms.
基金supported by a JSPS Grantin-Aid for Early-Career Scientists(JP22K17907).
摘要This study investigates the impact of the invariance of feature vectors for partial-to-partial point set registration under translation and rotation of input point sets,particularly in the realm of techniques based on deep learning and Gaussian mixture models(GMMs).We reveal both theoretical and practical problems associated with such deeplearning-based registration methods using GMMs,with a particular focus on the limitations of DeepGMR,a pioneering study in this line,to the partial-topartial point set registration.Our primary goal is to uncover the causes behind such methods and propose a comprehensible solution for that.To address this,we introduce an attention-based reference point shifting(ARPS)layer,which robustly identifies a common reference point of two partial point sets,thereby acquiring transformation-invariant features.The ARPS layer employs a well-studied attention module to find a common reference point rather than the overlap region.Owing to this,it significantly enhances the performance of DeepGMR and its recent variant,UGMMReg.Furthermore,these extension models outperform even prior deep learning methods using attention blocks and Transformer to extract the overlap region or common reference points.We believe these findings provide deeper insights into registration methods using deep learning and GMMs.Our source code and datasets are available at http://gffzzd16a119ca6e245c5su6kcq0q6v6w660un.ffgz.tsg.suse.edu.cn/DGRM-ARPS.git.
摘要在伪造音频检测任务中,高斯混合模型(GMM)虽然可以估计每个音频帧的概率分数,但未能充分利用高斯分量的概率分布信息,影响了检测性能。为此,本研究提出了一种基于联合特征建模与创新网络架构的双分支伪造音频检测方法。首先,结合线性频率倒谱系数(LFCC)和逆梅尔频率倒谱系数(IMFCC),设计了两种基于高斯概率分数的特征:L-GPF(LFCC-based Gaussian Probability Feature)和I-GPF(IMFCCbased Gaussian Probability Feature),分别捕捉音频的低频和高频特征。其次,提出了一种叠式双向长短时记忆网络(SBi-LSTM),分别对两类特征进行建模,从帧序列中学习真实与伪造音频的特征分布差异。最后,设计了创新的联合模块,通过动态权重融合和阈值因子判别机制整合双分支网络的特征信号,显著提高分类性能。实验在ASVspoof 2019数据集上进行。针对逻辑访问场景的开发集实验结果表明,相比于只使用I-GPF作为前端特征的单一SBi-LSTM模型,双分支模型的等错误率(EER)可降低94.6%,级联检测代价函数(t-DCF)可降低76.9%。相比于传统联合方式,在物理访问场景下联合模块最高可实现EER降低33.7%,t-DCF降低29.9%。相比于LFCC+GMM的基准模型,双分支模型的等错误率降低89.9%,级联检测代价函数降低89.2%。对比实验结果表明,与最新模型相比,双分支模型的EER降低15.6%,t-DCF降低29.8%。
基金funded by the Deanship of Scientific Research in Cooperation with Olive Research Center at Jouf University under Grant Number(DSR2022-RG-0163).
摘要Smart Agriculture,also known as Agricultural 5.0,is expected to be an integral part of our human lives to reduce the cost of agricultural inputs,increasing productivity and improving the quality of the final product.Indeed,the safety and ongoing maintenance of Smart Agriculture from cyber-attacks are vitally important.To provide more comprehensive protection against potential cyber-attacks,this paper proposes a new deep learning-based intrusion detection system for securing Smart Agriculture.The proposed Intrusion Detection System IDS,namely GMLPIDS,combines the feedforward neural network Multilayer Perceptron(MLP)and the Gaussian Mixture Model(GMM)that can better protect the Smart Agriculture system.GMLP-IDS is evaluated with the CIC-DDoS2019 dataset,which contains various Distributed Denial-of-Service(DDoS)attacks.The paper first uses the Pearson’s correlation coefficient approach to determine the correlation between the CIC-DDoS2019 dataset characteristics and their corresponding class labels.Then,the CIC-DDoS2019 dataset is divided randomly into two parts,i.e.,training and testing.75%of the data is used for training,and 25%is employed for testing.The performance of the newly proposed IDS has been compared to the traditional MLP model in terms of accuracy rating,loss rating,recall,and F1 score.Comparisons are handled on both binary and multi-class classification problems.The results revealed that the proposed GMLP-IDS system achieved more than 99.99%detection accuracy and a loss of 0.02%compared to traditional MLP.Furthermore,evaluation performance demonstrates that the proposed approach covers a more comprehensive range of security properties for Smart Agriculture and can be a promising solution for detecting unknown DDoS attacks.