变分自编码器模型综述

变分自编码器（VAE）作为深度隐空间生成模型的一种，近年来其表现性能取得了极大的成功，尤其是在图像生成方面。变分自编码器模型作为无监督式特征学习的重要工具之一，可以通过学习隐编码空间与数据生成空间的特征映射，进而在输出端重构生成输入数据。梳理了传统变分自编码器模型及其衍生变体模型的发展与研究现状，并就此做了总结和对比，最后分析了变分自编码器模型存在的问题与挑战，并就可能的发展趋势做了展望。     

Nonlinear modal analysis via non‐parametric machine learning tools

Modal analysis is an important tool in the structural dynamics community; it is widely utilised to understand and investigate the dynamical characteristics of linear structures. Many methods have been proposed in recent years regarding the extension to nonlinear analysis, such as nonlinear normal modes or the method of normal forms, with the main objective being to formulate a mathematical model of a nonlinear dynamical structure based on observations of input/output data from the dynamical system. In fact, for the majority of structures where the effect of nonlinearity becomes significant, nonlinear modal analysis is a necessity. The objective of the current paper is to demonstrate a machine learning approach to output‐only nonlinear modal decomposition using kernel independent component analysis and locally linear‐embedding analysis. The key element is to demonstrate a pattern recognition approach which exploits the idea of independence of principal components from the linear theory by learning the nonlinear manifold between the variables. In this work, the importance of output‐only modal analysis via “blind source” separation tools is highlighted as the excitation input/force is not needed and the method can be implemented directly via experimental data signals without worrying about the presence or not of specific nonlinearities in the structure.     

On the application of remote sensing towards the estimation of cultivated land lost to urbanization

In this research work, a 40-km² SPOT-5 High-Resolution Imagery (HRI) of the Warsak locality in district Peshawar, Pakistan, was utilized to approximate the quantity of cultivated land lost to urbanization, due to the construction of new homes and buildings. The imagery from a period of 2005 to 2015 for wheat crop was taken, specifically during the months of March and June when the crop is rich green and golden ripe respectively. eCognition ® program’s Object-Oriented Classification Method (OOCM) was employed for recognition of land versus buildings. Nearest Neighbour (NN), Support Vector Machine (SVM), Decision Trees (DT) and Random Forests (RF) were utilized for the classification process. The results demonstrated that the urbanized area had increased by approximately 28 per cent in the area considered. Moreover, the efficacy of the proposed method is depicted by an accuracy of 97.9 per cent and a Kappa Statistics of 0.975 for the SVM classifier.     

Multiple Empirical Kernel Learning with Majority Projection for imbalanced problems

Traditional Multiple Empirical Kernel Learning (MEKL) expands the expressions of the sample and brings better classification ability by using different empirical kernels to map the original data space into multiple kernel spaces. To make MEKL suit for the imbalanced problems, this paper introduces a weight matrix and a regularization term into MEKL. The weight matrix assigns high misclassification cost to the minority samples to balanced misclassification cost between minority and majority class. The regularization term named Majority Projection (MP) is used to make the classification hyperplane fit the distribution shape of majority samples and enlarge the between-class distance of minority and majority class. The contributions of this work are: (i) assigning high cost to minority samples to deal with imbalanced problems, (ii) introducing a new regularization term to concern the property of data distribution, (iii) and modifying the original PAC-Bayes bound to test the error upper bound of MEKL-MP. Through analyzing the experimental results, the proposed MEKL-MP is well suited to the imbalanced problems and has lower generalization risk in accordance with the value of PAC-Bayes bound.     

Tibetan Multi-Dialect Speech and Dialect Identity Recognition

Tibetan language has very limited resource for conventional automatic speech recognition so far. It lacks of enough data, sub-word unit, lexicons and word inventories for some dialects. And speech content recognition and dialect classification have been treated as two independent tasks and modeled respectively in most prior works. But the two tasks are highly correlated. In this paper, we present a multi-task WaveNet model to perform simultaneous Tibetan multi-dialect speech recognition and dialect identification. It avoids processing the pronunciation dictionary and word segmentation for new dialects, while, in the meantime, allows training speech recognition and dialect identification in a single model. The experimental results show our method can simultaneously recognize speech content for different Tibetan dialects and identify the dialect with high accuracy using a unified model. The dialect information used in output for training can improve multi-dialect speech recognition accuracy, and the low-resource dialects got higher speech content recognition rate and dialect classification accuracy by multi-dialect and multi-task recognition model than task-specific models.     

Couple-based particle swarm optimization for short-term hydrothermal scheduling

A novel couple-based particle swarm optimization (CPSO) is presented in this paper, and applied to solve the short-term hydrothermal scheduling (STHS) problem. In CPSO, three improvements are proposed compared to the canonical particle swarm optimization, aimed at overcoming the premature convergence problem. Dynamic particle couples, a unique sub-group structure in maintaining population diversity, is adopted as the population topology, in which every two particles compose a particle couple randomly in each iteration. Based on this topology, an intersectional learning strategy using the partner learning information of last iteration is employed in every particle couple, which can automatically reveal useful history information and reduce the overly rapid evolution speed. Meanwhile, the coefficients of each particle in a particle couple are set as distinct so that the particle movement patterns can be described and controlled more precisely. In order to demonstrate the effectiveness of our proposed CPSO, the algorithm is firstly tested with four multimodal benchmark functions, and then applied to solve an engineering multimodal problem known as STHS, in which two typical test systems with four different cases are tested, and the results are compared with those of other evolutionary methods published in the literature.     

Vibration analysis and identification of breathing cracks in beams subjected to single or multiple moving mass using online sequential extreme learning machine

In this paper, a novel approach was presented to vibration analysis and identification of breathing cracks in Timoshenko beam under single or multiple moving mass. Dynamic strain energies (DSEs) and translational accelerations in beam structures under moving mass were used as forward problem and application of an emergent learning algorithm called the online sequential extreme learning machine algorithm as inverse problem to predict crack depths and locations. To demonstrate the potential of the proposed vibration analysis over existing ones, two validation studies have been done. To evaluate the proposed method to identify breathing cracks, two examples, namely, clamped–clamped beam and two span continuous beams have been studied. Also, the effect of the discrepancy in stiffness between the finite-element model and the actual tested dynamic system has been investigated. Another examination has been performed in which moving mass with different speeds are utilized. Also, the effect of multi mass passing through the beam has been studied. The obtained results indicated that the proposed method could identify the breathing cracks existence and severity in the beam under moving mass using DSE and accelerations, which may be noisy or noise free.     

基于深度学习的四旋翼无人机单目视觉避障方法

针对无人机避障问题，提出一种基于深度学习的四旋翼无人机单目视觉避障方法。首先通过目标检测框选出目标在图像中的位置，并通过计算目标选框上下边距的长度，以此来估量出障碍物到无人机之间的距离；然后通过协同计算机判断是否执行避障动作；最后使用基于Pixhawk搭建的飞行实验平台进行实验。实验结果表明，该方法可用于无人机低速飞行条件下避障。该方法所用到的传感器只有一块单目摄像头，而且相对于传统的主动式传感器避障方法，所占用无人机的体积大幅减小。该方法鲁棒性较好，能够准确识别不同姿态的人，实现对人避障。