基于Vague-Fuzzy理论的深基坑支护方案评价应用

深基坑支护决策是典型的非结构或半结构决策问题,对其进行量化研究对于实现工程方案的科学决策具有重要的理论意义与实践价值.采用基于Vague-Fuzzy理论的深基坑支护决策评价方法.在研究评价指标Vague值确定方法的基础上,建立了Vague集近似转化为Fuzzy集的方法与模型,得到的模糊隶属度并可根据参数k的数学性质有目的地进行调整,从而实现深基坑支护决策的模糊评价.以某大厦深基坑支护拟采用的4种方案评价为例,对所建模型进行了实例研究,结果表明该方法是科学、实用的.     

深度残差收缩网络的含噪微泄漏超声识别方法*

在利用声学信号进行泄漏检测时,复杂的背景噪声往往会淹没微弱的泄漏信号,导致误判率高。针对微小泄漏在含噪环境中识别困难的问题,提出了基于深度残差收缩网络（DRSN）的含噪微泄漏识别方法。在提出的方法中,添加不同强度高斯噪声,建立数据集,使用DRSN网络进行训练,验证DRSN对不同泄漏强度、不同噪声含量样本识别的有效性。实验结果表明：DRSN对于微弱泄漏可以达到较理想的识别率,即使在高度杂糅数据识别时仍能达到较理想的识别效果,而且噪声含量并不会对DRSN迭代次数产生明显的影响。将提出的方法与CNN识别方法对比,DRSN具有明显的优势。     

基于变分推断的磁共振图像群组配准

为解决基于深度学习的成对配准方法精度低和传统配准算法耗时长的问题,本文提出一种基于变分推断的无监督端到端的群组配准以及基于局部归一化互相关（NCC）和先验的配准框架,该框架能够将多个图像配准到公共空间并有效地控制变形场的正则化,且不需要真实的变形场和参考图像．该方法得到的预估变形场可建模为概率生成模型,使用变分推断的方法求解;然后借助空间转换网络和损失函数来实现无监督方式训练．对于公开数据集LPBA40的3D脑磁共振图像配准任务,测试结果表明：本文所提出的方法与基线方法相比,具有较好的Dice得分、运行时间少且产生更好的微分同胚域,同时对噪声具有鲁棒性．     

基于深度神经网络的时空编码磁共振成像超分辨率重建方法

单扫描时空编码磁共振成像是一种新型超快速磁共振成像技术,它对磁场不均匀和化学位移伪影有较强的抵抗性,但是其固有的空间分辨率较低,因此通常需要进行超分辨率重建,以在不增加采样点数的情况下提高时空编码磁共振图像的空间分辨率.然而,现有的重建方法存在迭代求解时间长、重建结果有混叠伪影残留等问题.为此,本文提出了一种基于深度神经网络的单扫描时空编码磁共振成像超分辨率重建方法.该方法采用模拟样本训练深度神经网络,再利用训练好的网络模型对实际采样信号进行重建.数值模拟、水模和活体鼠脑的实验结果表明,该方法能快速重建出无残留混叠伪影、纹理信息清楚的超分辨率时空编码磁共振图像.适当增加训练样本数量以及在训练样本中加入适当的随机噪声水平,有助于改善重建效果.     

Deep learning for image reconstruction in thermoacoustic tomography

Microwave-induced thermoacoustic tomography(TAT)is a rapidly-developing noninvasive imaging technique that integrates the advantages of microwave imaging and ultrasound imaging.While an image reconstruction algorithm is critical for the TAT,current reconstruction methods often creates significant artifacts and are computationally costly.In this work,we propose a deep learning-based end-to-end image reconstruction method to achieve the direct reconstruction from the sinogram data to the initial pressure density image.We design a new network architecture TAT-Net to transfer the sinogram domain to the image domain with high accuracy.For the scenarios where realistic training data are scarce or unavailable,we use the finite element method(FEM)to generate synthetic data where the domain gap between the synthetic and realistic data is resolved through the signal processing method.The TAT-Net trained with synthetic data is evaluated through both simulations and phantom experiments and achieves competitive performance in artifact removal and robustness.Compared with other state-of-the-art reconstruction methods,the TAT-Net method can reduce the root mean square error to 0.0143,and increase the structure similarity and peak signal-to-noise ratio to 0.988 and 38.64,respectively.The results obtained indicate that the TAT-Net has great potential applications in improving image reconstruction quality and fast quantitative reconstruction.     

Acoustic multipath structure in direct zone of deep water and bearing estimation of tow ship noise of towed line array

In the towed line array sonar system, the tow ship noise is the main factor that affects the sonar performance. Conventional noise cancelling methods assume that the noise is towards the endfire direction of the array. An acoustic experiment employing a towed line array is conducted in the western Pacific Ocean, and a strange bearing-splitting phenomenon of the tow ship noise is observed in the array. The tow ship noise is split into multiple noise signals whose bearings are distributed between 10° and 90° deviating from the endfire direction. The multiple interferences increase the difficulty in recognizing the target for the sonar operator and noise cancellation. Therefore, making the mechanism clear and putting forward the tow ship noise splitting bearing estimation method are imperative. In this paper, the acoustic multi-path structure of the tow ship in deep water is analyzed. Then it is pointed out that the bearing-splitting phenomenon is caused by the main lobe of direct rays and bottom-reflected rays, as well as several side lobes of direct rays. Meanwhile, the indistinguishability between the elevation angle and the bearing angle due to the axial symmetry of a strict horizontal line array causes the bearing to deviate from the endfire direction. Based on the theory above, a method of estimating bearing of the tow ship noise in deep water is proposed. The theoretical analysis results accord with the experimental results, which helps to identify the target and provide correct initial bearing guidance for noise cancelation methods.     

Deep-learning-based cryptanalysis of two types of nonlinear optical cryptosystems

The two types of nonlinear optical cryptosystems (NOCs) that are respectively based on amplitude-phase retrieval algorithm (APRA) and phase retrieval algorithm (PRA) have attracted a lot of attention due to their unique mechanism of encryption process and remarkable ability to resist common attacks. In this paper, the securities of the two types of NOCs are evaluated by using a deep-learning (DL) method, where an end-to-end densely connected convolutional network (DenseNet) model for cryptanalysis is developed. The proposed DL-based method is able to retrieve unknown plaintexts from the given ciphertexts by using the trained DenseNet model without prior knowledge of any public or private key. The results of numerical experiments with the DenseNet model clearly demonstrate the validity and good performance of the proposed the DL-based attack on NOCs.     

Experience-driven learning-based intelligent hybrid beamforming for massive MIMO mmWave communications

We have studied massive MIMO hybrid beamforming (HBF) for millimeter-wave (mmWave) communications, where the transceivers only have a few radio frequency chain (RFC) numbers compared to the number of antenna elements. We propose a hybrid beamforming design to improve the system’s spectral, hardware, and computational efficiencies, where finding the precoding and combining matrices are formulated as optimization problems with practical constraints. The series of analog phase shifters creates a unit modulus constraint, making this problem non-convex and subsequently incurring unaffordable computational complexity. Advanced deep reinforcement learning techniques effectively handle non-convex problems in many domains; therefore, we have transformed this non-convex hybrid beamforming optimization problem using a reinforcement learning framework. These frameworks are solved using advanced deep reinforcement learning techniques implemented with experience replay schemes to maximize the spectral and learning efficiencies in highly uncertain wireless environments. We developed a twin-delayed deep deterministic (TD3) policy gradient-based hybrid beamforming scheme to overcome Q-learning’s substantial overestimation. We assumed a complete channel state information (CSI) to design our beamformers and then challenged this assumption by proposing a deep reinforcement learning-based channel estimation method. We reduced hybrid beamforming complexity using soft target double deep Q-learning to exploit mmWave channel sparsity. This method allowed us to construct the analog precoder by selecting channel dominant paths. We have demonstrated that the proposed approaches improve the system’s spectral and learning efficiencies compared to prior studies. We have also demonstrated that deep reinforcement learning is a versatile technique that can unleash the power of massive MIMO hybrid beamforming in mmWave systems for next-generation wireless communication.     

Revisiting the dynamics of Bose–Einstein condensates in a double well by deep learning with a hybrid network

Deep learning, accounting for the use of an elaborate neural network, has recently been developed as an efficient and powerful tool to solve diverse problems in physics and other sciences. In the present work, we propose a novel learning method based on a hybrid network integrating two different kinds of neural networks: Long Short-Term Memory (LSTM) and Deep Residual Network (ResNet), in order to overcome the difficulty met in numerically simulating strongly-oscillating dynamical evolutions of physical systems. By taking the dynamics of Bose–Einstein condensates in a double-well potential as an example, we show that our new method makes a highly efficient pre-learning and a high-fidelity prediction about the whole dynamics. This benefits from the advantage of the combination of the LSTM and the ResNet and is impossibly achieved with a single network in the case of direct learning. Our method can be applied for simulating complex cooperative dynamics in a system with fast multiplefrequency oscillations with the aid of auxiliary spectrum analysis.     

图嵌入算法研究进展

图嵌入算法是将高维网络信息映射至低维后用实数向量表示的一种方法,用于解决推荐系统、社区发现及节点分类等。近年来,随着科技的进步,图数据呈现海量、异构、高维、多模态等特点,机器学习等人工智能算法对高性能的图嵌入算法的需求日益增加,图嵌入已成为国内外人工智能领域的研究热点之一。对图嵌入算法的研究进展、技术原理及基础理论进行了综述,系统概述了已有的主流图嵌入算法,包括基于降维方法的图嵌入、基于矩阵分解的图嵌入、基于网络拓扑结构的图嵌入、基于神经网络的图嵌入、基于生成式对抗网络的图嵌入和基于超图网络的图嵌入,对这些算法进行了分析与比较,并给出了相应的应用场景;归纳总结了常用的测试数据集及其评价标准;最后,展望了图嵌入算法的研究趋势和方向。