关键词： Behavioral Health   Depression   Machine Learning   Prevention  

摘要： Behavioral health disorders, specifically depression, are a serious health concern in the United States and worldwide. The consequences of unaddressed behavioral health conditions are multifaceted and have impact at the individual, relational, communal, and societal level. Despite the number of individuals who could benefit from treatment for behavioral health concerns, their difficulties are often unidentified and unaddressed through treatment. Technology carries unrealized potential to identify people at risk for behavioral health conditions and to inform prevention and intervention strategies. Drawing upon data from the National Longitudinal Study of Adolescent Health (Add Health, n=3782), this study has two aims related to advancing understanding of technology’s potential value in behavioral health: 1) to develop a forecasting procedure that can be used to identify youth who are at risk of reporting a depression diagnosis as adults based on a set of input variables; and 2) to understand the developmental trajectories of depression for youth. To address the first aim of this study, random forest methodology was used to derive the forecasting algorithm. The second aim was pursued with Generalized Additive Model analysis to estimate relationships between presence of a reported depression diagnosis as an adult and youth characteristics. Findings from this study indicate that it is feasible to use a forecasting tool to identify individuals at risk of being diagnosed with depression, which can facilitate early intervention and improved outcomes. Gender, race, and receiving counseling as a youth were the most important predictors of having a reported depression diagnosis as an adult. This dissertation addresses the role of health disparities, specifically gender and race, related to depression and mental health treatment. In sum, this dissertation highlights how a machine learning forecasting tool could be used to inform prevention strategies and understanding of facto

关键词： 机器学习   分类算法   SVM   逻辑回归   随机森   KNN-SVM  

摘要： 随着数据量指数级别的增长和计算机硬件设备的不断完善,机器学习算法迎来了发展的黄金时代。以熟知的互联网公司为例,百度的语音搜索、科大讯飞的语音识别、字节跳动的个性化推荐算法和滴滴出行的智能调度算法,都是机器学习算法在日常生活中的应用。分类算法作为机器学习算法中最常见的算法,有着坚实的数学理论基础,为大数据的处理和人工智能的发展提供了有效的方法。本文对机器学习中经典分类算法的研究背景和意义进行了阐述,并对机器学习中分类算法的国内外研究现状进行了总结。其中,重点介绍了四种经典的监督学习分类算法(逻辑回归、朴素贝叶斯、SVM和决策树)和一种集成学习的分类算法(随机森林),并详细地研究和分析了它们背后的数学原理。本文利用20世纪90年代美国人口普查数据,对居民年收入是否大于50k$进行分类预测。首先,利用数据类型转换、缺失值处理和数据标准化等方法,对数据进行预处理;其次使用三种经典的监督学习分类算法(逻辑回归、朴素贝叶斯和SVM)和一种集成学习的分类算法(随机森林)对预处理后的数据进行分类,并且使用混淆矩阵、准确率、精确率、召回率、F1值和ROC曲线对分类结果进行评价;最后基于SVM算法在分类超平面附近容易错分的缺点,提出将SVM算法和KNN算法相结合的组合算法,同时KNN-SVM算法再根据KNN样本点的选取规则的不同分成KSVM1和KSVM2两种算法,通过实验对比,两种新算法都比单纯使用SVM算法在分类准确率上有所提高。

关键词： Industrial engineering   Computer science  

摘要： The rise of social media networks such as Twitter presents us with a massively distributed portal of information covering every topic from social and political news to entertainment. However, this coverage of different topics coupled with the high rate of content generation on Twitter results in an overwhelming volume of content that makes it difficult to detect and retrieve information relevant to a specific topic. Access to efficient methods for retrieval and analysis of this topical content can be of great value in a lot of different domains from social sciences to marketing analysis. Hence, the concept of learning topical classifiers and predictors for Twitter has become an important area of research. This thesis aims to address the challenges of topical content retrieval and analysis via efficient ways that will allow users to have real-time access to insights on their desired topics.

关键词： 结构健康监测   模态参数识别   机器学习   深度网络   模态独立性  

摘要： 结构的模态参数(频率、振型、阻尼比)是结构的一个固有本质特性,是了解结构动力特性、进行结构损伤检测与安全评估的重要依据。然而,传统的结构模态参数识别方法在应用时往往需要较多的人工干预和参数设置,在结构健康监测系统中难以实现模态参数的自动识别,严重影响结构健康监测系统在线预警。如何实现结构模态参数的智能识别是值得研究的问题,为此,本文结合机器学习理论,研究环境激励下结构模态参数的智能方法。本文主要研究内容包括:提出了一种基于模态独立性的环境激励下结构模态参数识别的机器学习方法。该方法首先将模态参数识别问题转化为深度学习问题,然后考虑模态响应的不相关性和非高斯性构造损失函数以估算模态参数。对数值模拟算例和某斜拉桥结构健康监测数据进行了验证,结果表明,所提出的方法与常用的频域分解法,随机子空间法和NEx T+ERA方法等具有相似的模态识别精度,但明显优于传统独立成分分析方法。提出的方法具有一定的智能数据处理能力,为环境激励下的结构模态参数识别提供一种新的选择。提出了一种基于结构响应时频域稀疏性的环境激励下结构模态参数识别的机器学习方法。利用结构振动响应数据在时频域具有稀疏性的特点,首先将结构振动响应数据从时域变换到时频域,然后挑选出对应于各阶模态处的响应单源点,其次对单源点进行对称化和归一化,最后利用设计的竞争型神经网络求得单源点的聚类中心,也就是各阶振型系数的值。对某斜拉桥实际结构健康监测数据进行了验证,所提方法能较好的识别得到振型系数。提出了一种基于有监督多任务学习的环境激励下结构模态参数识别的机器学习方法。该方法将结构系统响应数据作为网络输入,分离得到的各阶模态响应作为标签,通过发挥深度神经网络强大的学习能力,训练深度网络并使其能自动将系统响应分离为模态响应,得到结构模态响应和振型系数。通过数值模拟算例验证了方法的有效性。

关键词： 预测电力输出   联合循环发电厂   机器学习方法   REPTree装袋算法   最佳子集   SK-BREP算法  

摘要： 预测联合循环发电厂的满载电力输出是一项重要工作,能够直接决定电厂发电量,并使其获取最大利益。如果要用热力学方式准确分析系统,需要对大量运行条件进行设定以解释公式中的不确定参数。如果没有这些设定,一个真实应用的热力学分析需要数千个非线性方程,尤其是在分析任意的输入和输出模式的系统中,直接进行求解具有较高难度。为了解决上述问题,本文提出采用机器学习的方法替代热力学方式的求解。电厂的基载运行主要受4个参数影响,它们是环境温度,气压,相对湿度,废气压力,作为数据集的输入变量,这些参数影响电力输出,电力输出便作为目标变量。为了比较这些参数的影响,对参数的组合形式进行了测试,在所有特征子集中寻找最佳数据子集,本文使用的包含输入变量和目标变量的数据集,通过实验测试,确定了使得机器学习算法具有最佳性能的输入变量组合形式。本论文针对不同机器学习方法进行测试与比较,确定REP Tree装袋算法预测效果是最好的。针对该算法对参量的适应性不高的问题,进一步提出SK-BREP算法,利用k-means聚类的思想,将所训练出来的分类回归树按预测结果的相似度进行聚类,然后从每一类中再选择出预测精度高的一批分类回归树参加集成。基于朝阳发电厂的运行数据进行了测试,结果表明本文所提出的SK-BREP算法能够实现对联合循环发电厂每小时满载电力输出的准确预测。该论文有图10幅,表10个,参考文献46篇,对所研究的过程和结论进行阐述和分析。

关键词： 地下水资源量   动态评价模型   静态要素   动态要素   遥感大数据   多次BP神经网络学习   自然分区   实时变化监测  

摘要： 地下水资源量表示地下水饱和含水层逐年更新的动态水量,即降水和地表水入渗对地下水的补给量。对地下水资源量进行评价,有利于为地下水资源保护和合理配置提供依据,有利于为合理高效开发地下水资源提供重要的技术支撑,有利于减少不合理开发利用地下水引起的生态及环境地质问题,还有利于经济社会的可持续发展和生态环境建设。利用遥感数据反演地下水资源量时,由于遥感数据属于大数据,具有不同的时间、空间尺度,而作为真值数据的地下水资源量公报数据获取成本高、周期长,因此数据量十分有限,这导致遥感数据与地下水资源量公报数据之间尺度不匹配。常用的解决数据尺度不匹配问题的方法是降低遥感数据尺度到与公报数据尺度相同的水平。但这又会导致遥感数据利用不够充分,且生成的模型的精度受到限制。基于此,本文提出了一种建立地下水资源量动态评价模型的方法,该方法保持遥感数据尺度不变,能够有效地匹配不同尺度的数据,并采用机器学习方法揭示遥感数据与地下水资源量公报数据之间的关联,从而实现对地下水资源量的动态评价。文中以广东省为例,建立了该地区的地下水资源量动态评价模型,并对该模型进行精度检验,且利用该模型实现了广东省以地下水资源量分级标准为依据的自然分区的划分,以及广东省短时间间隔的地下水资源量的实时变化监测。本文主要取得了如下成果和认识:1、本文提出基于遥感大数据和机器学习方法建立地下水资源量动态评价模型的方法,并利用该方法建立广东省地下水资源量动态评价模型,该模型具有较好的精度。研究中利用多源(MODIS、TRMM、Landsat ETM+/OLI、SRTM等数据)、多时相(2004-2015年)、多尺度的遥感数据及图文资料提取与地下水相关要素(按其随时间变化的频率分为动态要素和静态要素)的栅格数据。随后对地下水资源量公报数据进行栅格化处理,使之与要素数据具有相同的坐标系和空间分辨率,以获取大数据学习样本。然后将学习样本输入神经网络进行多次学习,每次学习完成后剔除部分误差较大的学习样本,最终利用比较符合规律的学习样本建立广东省地下水资源量动态评价模型。文中利用两个级别的行政分区(地级市和区县)的地下水资源量公报数据对模型进行检验,检验结果表明模型具有较好的精度:大行政分区(地级市)的地下水资源量公报数据共252个,其中相对误差小于20%的数据占66.4%,小于30%的数据占79.5%;小行政分区(区县)的地下水资源量公报数据共235个,其中相对误差小于20%的数据占60.9%,小于30%的数据占79.6%。2、本文提出的建模方法通过对地下水资源量公报数据进行栅格化处理,解决了要素栅格数据与地下水资源量公报数据之间尺度不匹配的问题。为了提高模型精度,在每次学习完成后剔除部分误差较大的样本,并经过多次学习,最终利用比较符合规律的学习样本建立地下水资源量动态评价模型。采用常用方法解决数据尺度不匹配的问题需要降低要素数据尺度到与地下水资源量公报数据尺度相同的水平。基于常用方法,学习样本数量有限(仅252个),利用这些样本建立的地下水资源量动态评价模型比利用本研究方法建立的地下水资源量动态评价模型的精度更低。3、利用地下水资源量动态评价模型计算广东省各行政区中各个像元的地下水资源量值,该值可以反映各行政区域内部地下水资源量不均匀分布的特征,弥补了一个地下水资源量公报数据表示一个区域的地下水资源量的整体状况的不足。4、利用地下水资源量动态评价模型计算广东省的地下水资源量,并制定地下水资源量分级标准,成功划定研究区以地下水资源量分级标准为依据的自然分区(文中以2007年和2009年为例)。这种自然分区比行政分区能更好地反映各个要素与地下水资源量的关系,能够更客观地评价地下水资源量,从而为各种科学研究提供更好的支撑。5、利用地下水资源量动态评价模型可以实现短时间间隔的地下水资源量动态变化监测。研究利用地下水资源量动态评价模型计算广东省三个时间段内不同时相(以天为间隔)的地下水资源量,并对各时段内地下水资源量的变化情况进行分析。

关键词： Mental health   Behavioral psychology   Computer science  

摘要： Predictive analytics and computational phenotyping techniques have shown promising results in several areas of medicine, including automated classification of radiology imaging, survival analysis in cancer patients, and prediction of life-threatening events in hospitalized patients. In recent years, computational psychiatry has emerged as a field that combines multiple levels and types of data and computational model- ing to improve understanding, prediction, and treatment of mental illness. Mental health patients often undergo a variety of non-invasive (e.g., cognitive counseling) and invasive (e.g., surgery) therapies before finding an effective treatment plan. Im- proved prediction of treatment response can shorten the duration of clinical trials and improve patient experience and outcomes. A key challenge of applying predictive modeling to this problem is that often, the effectiveness of a treatment regimen remains unknown for several weeks. In this thesis, we propose Machine Learning approaches to extracting audio-visual features for predicting the likely outcome of Deep Brain Stimulation (DBS) treatment several weeks in advance for patients suffering from major depressive disorder, a common psychiatric illness for which there are no objective, non-verbal, automated markers that can reliably track treatment response. We first explore the use of video analysis of facial expressivity in a cohort of severely depressed patients before and after DBS. We introduce a set of variability measurements to obtain unsupervised features from muted video recordings. We then leverage the link between short-term emotions and long-term depressed mood states and use a neural network model on the top of emotion-based audio features. The results show that unsupervised features extracted from these audio and video recordings, when incorporated in classification models, can discriminate different lev- els of depression severity during ongoing DBS treatment. Moreover, for the long ter

关键词： 2D Materials   Ferroelectric Rashba   Graphene   Machine Learning   Valleytronics  

摘要： Monolayers and heterostructures of two-dimensional (2D) electronic materials with spin-orbit interactions offer the promise of observing many novel physical effects. While theoretical predictions of 2D layered materials based on density functional theory (DFT) are many, the DFT approach is limited to small simulation sizes (several nanometers), and thus inhomogeneous strain and boundary effects that are often observed experimentally cannot be simulated within a reasonable time. The aim of this thesis is (i) to study effects of strain on 2D materials beyond graphene using first-principles and tight-binding methods and (ii) to investigate the effects of cuts–"kirigami"– on 2D materials using molecular dynamics and machine learning approach. The first half of this thesis focuses on the effects of strain on manipulating spin and valley degrees of freedom for two classes of 2D materials–monochalcogenide and lead chalcogenide monolayers–using DFT. A tight-binding (TB) approach is developed to describe the electronic changes in lead chalcogenide monolayers due to strains that often persist in real devices. The strain-dependent TB model allows one to establish a relationship between the Rashba field and the out-of-plane strain or electric polarization from a microscopic view, a connection that is not well understood in the ferroelectric Rashba materials. This framework connecting strain fields and electronic changes is important to overcome the size and computational limitations associated with DFT. The second part of the thesis focuses on defect engineering and design of 2D materials via the "kirigami" technique of introducing different patterns of cuts. A machine learning (ML) approach is presented to provide physical insights and an effective model to describe the physical system. We demonstrate that a machine learning model based on a convolutional neural network is able to find the optimal design from a training data set that is much smaller than the design space.

关键词： 航空发动机   故障诊断   传感器解析余度   性能衰退预测   极限学习机  

摘要： 航空发动机故障诊断与性能预测技术是健康管理系统(PHM)的重要组成部分,是保证飞行安全、降低维护成本的重要手段。本文以涡扇发动机为主要研究对象,研究了基于极限学习机和深度学习算法的航空发动机气路故障诊断以及性能预测方法,主要工作内容如下:为提高极限学习机(ELM)计算结果稳定性,采用基于受限玻尔兹曼机(RBM)的ELM参数初始化方法,设计了改进的ELM学习算法(RB-ELM)。该方法通过RB初始化过程来优化ELM初始参数,降低了ELM网络中存在的参数随机性以及由此造成的学习性能不稳定。将RB-ELM算法应用到航空发动机气路故障诊断中,并验证了在飞行包线内典型工况下具有较好分类精度。进一步研究了在线序列极限学习机(OS-ELM)算法,发展了在线学习稳定性提高策略。基于OS-ELM的集成学习策略(EOS-ELM),研究了一种基于线性卡尔曼滤波的自适应集成学习机KEOS-ELM,该算法对每个集成的OS-ELM网络的学习误差进行评价,通过滤波算法调整权重。将KEOS-ELM算法应用于航空发动机传感器解析余度,改进了传感器漂移故障的判别方法并进行了数字仿真验证。将线性卡尔曼滤波算法用于OS-ELM网络权重的迭代更新,提出了一种OS-ELM的线性卡尔曼滤波训练方法(KFOS-ELM),相比于传统递推最小二乘求解训练方法,提高了OS-ELM学习过程的稳定性,并将KFOS-ELM算法与逻辑回归模型(Logisic-Regression)综合,研究了一种航空发动机性能衰退预测方法,KFOS-ELM作为参数预测器,以逻辑回归模型作为退化状态评价模型,实现了发动机失效时间的短期预测。最后,研究了基于深度学习算法长短期记忆网络(LSTM)和自回归积分滑动模型(ARIMA)的发动机剩余使用寿命预测方法。该方法基于发动机历史退化数据建立健康指标模型库,通过ARIMA模型得到的预测参数解算健康指标,得到其剩余使用寿命以及拟合的概率分布结果,实现了发动机剩余使用寿命预测。

关键词： Applied mathematics   Artificial intelligence  

摘要： This dissertation presents novel approaches and applications of machine learning architectures. In particular, these approaches are based on tools from topological data analysis and are used in conjunction with conventional machine learning methods. Topological data analysis, which is based on algebraic topology, can identify significant global mathematical structures which are out of reach of many other approaches. When we use topology we benefit from generality, and when we use conventional methods we benefit from specificity. This dissertation contains a broad overview of data science and topological data analysis, then transitions to three distinct machine learning applications of these methods. The first application uses linear methods to discover the inherent dimensionality of the manifold given by congressional roll call votes. The second uses persistent homology to identify extremely noisy images in both supervised and unsupervised tasks. The last application uses mapper objects to produce robust classification algorithms. Two additional projects are presented later in the appendix, and are related to the three main applications. The first of these constructs a method to choose optimal optimizers, and the second places mathematical constraints on the structure of renormalization group flows.