关键词： Statistics   Bayesian   categorical predictor   classification   mixture model   prediction   Statistics and Probability  

摘要： A variety of conditional probability models estimate the regression or class probability function for the purpose of prediction or classification. Bayesian mixture models provide flexible prediction and classification methods for modeling local linearities of the regression or class probability function. A hierarchical Bayes Gaussian mixture model is proposed that directly uses data to define a mixture prior for its Gaussian mixture component parameters. This nonparametric Bayesian mixture model uses the stick-breaking construction of a Dirichlet process model. Prediction and classification comes directly from the posterior distribution via Gibbs sampling. Comprehensive simulation studies demonstrate performance of both the regression and classification methods. Five standard machine learning data sets show prediction and classification results competitive with local methods. A generic classification algorithm is outlined given categorical predictors. If too many categories are present or if many interaction levels affect the class probability function, no current methods can reduce bias effectively. A proposed solution is a generic way to characterize the information about the class probability function available in the predictors through likelihood ratio statistics. This proposed classifier relies on random forests to reduce bias by utilizing all information in the generated log likelihood ratio features. A simulation study and an application data set demonstrate potential advantages of this classification method for categorical predictors.

关键词： P-glycoprotein   ADMET   multidrug resistance   QSAR   data mining  

摘要： P-glycoprotein (Pgp) is a drug transporter that plays important roles in multidrug resistance and drug pharmacokinetics. The inhibition of Pgp has become a notable strategy for combating multidrug-resistant cancers and improving therapeutic outcomes. However, the polyspecific nature of Pgp, together with inconsistent results in experimental assays, renders the determination of endpoints for Pgp-interacting compounds a great challenge. In this study, the classification of a large set of 2,477 Pgp-interacting compounds (i.e., 1341 inhibitors, 913 non-inhibitors, 197 substrates and 26 non-substrates) was performed using several machine learning methods (i.e., decision tree induction, artificial neural network modelling and support vector machine) as a function of their physicochemical properties. The models provided good predictive performance, producing MCC values in the range of 0.739-1 for internal cross-validation and 0.665-1 for external validation. The study provided simple and interpretable models for important properties that influence the activity of Pgp-interacting compounds, which are potentially beneficial for screening and rational design of Pgp inhibitors that are of clinical importance.

关键词： 生物特征识别   机器学习   未匹配细节点   异源人脸相似度   序列化多模态生物特征识别  

摘要： 生物特征识别技术作为一种新型的身份认证手段,其研究价值和良好的应用前景得到了学术界和企业界的公认。自上世纪九十年代中期以来,经国内外学术界和企业界同行的多年推进,生物特征识别的研究取得了巨大进展,已有较多的生物特征识别技术在诸多领域得到有效应用。但是,现有生物特征识别系统性能上的不完善性,在很大程度上影响了生物特征识别技术的推广应用。如何进一步提高生物特征识别系统的性能,使其能够更好地满足实际应用的需求,仍是富有挑战性的、值得深入研究的问题。近年来,机器学习技术成为计算机领域中的研究热点,并且不断发展,在很多方面得到成功应用。在生物特征识别研究领域,利用机器学习相关方法解决问题已成为一个主流的趋势。例如利用集成学习思想成功解决多模板集成问题,利用半监督学习思想成功处理生物特征的类内差异性问题等。本文深入分析目前在指纹识别、人脸识别及多模态生物特征识别领域影响生物特征识别系统性能及其应用的几个关键问题,设计并运用合理的机器学习技术解决这些问题,从而提高了人脸识别、指纹识别系统的识别准确率,并且获得了多模态生物识别系统使用方便性和识别准确率的同时提升。本文的主要工作和贡献有：1、基于细节点的指纹匹配一直是指纹识别中最为主流的方法。迄今为止,基于细节点的指纹匹配中,未能形成匹配的细节点都被视为无用信息。实际上,由于同源指纹和异源指纹匹配中未匹配细节点的成因和特点不同,未匹配细节点中包含一定的区分性信息。如何合理挖掘和利用这些信息,是提升自动指纹识别系统性能的一个关键问题。基于此分析,本文定义和提取7个未匹配细节点特征作为辅助特征,获取7个辅助匹配得分,并利用集成学习的方法,将辅助匹配得分与传统匹配得分进行融合,实现指纹识别。在FVC2000、FVC2002和FVC2004国际指纹识别竞赛数据库上验证了未匹配细节点特征的区分能力和所提方法在提高系统识别性能方面的有效性。在3个竞赛数据库所包含的12个子数据库上,本文方法将等错误率(Equal Error Rate)最少降低了33.6%,最多降低了77.0%。2、在目前的人脸识别技术中,判断两个人脸样本是否属于同一个体,只利用了这两个人脸样本的相似度信息,而忽略了其他的可用信息。经分析,取自不同个体的人脸样本之间的异源相似度信息具有稳定规律,且少数显著异源相似度信息具有好的区分能力,如何合理挖掘和利用这些信息并运用于人脸识别是提高人脸识别系统识别性能的关键问题。基于此,本文提出基于稀疏相似度序列的人脸识别方法(S3M),用相似度序列定义和提取异源相似度信息,并利用Lasso稀疏学习方法提取其中少数显著信息,在此过程中,提出区分性判定准则优化稀疏参数,最后利用稀疏相似度序列作为高一级特征进行人脸识别。经实验验证,S3M在两个公开人脸数据库上等错误率分别达到了0.05%和0.01%,有效提升了系统的识别性能。3、现有的多模态生物特征识别系统主要采用并行集成模式,在注册和识别阶段均需采用系统内所有生物特征进行识别,给用户带来极大不便。在使用方便性和效率方面的这一缺点限制了多模态生物特征识别系统的广泛应用。本文分析并提出最符合广大用户使用需求的多模态生物特征识别系统应采用序列化集成框架,将采集(使用)方便的特征放在序列前端,而尽量减少对采集不方便特征的使用。一切对系统性能的优化应该在这个框架下进行。然而,采集方便的生物特征往往识别性能较低,特征的使用方便性和识别准确率存在矛盾。这一矛盾问题下,为保证系统的识别准确性,仍有大量用户需要使用方便性较差的特征进行识别,抵消了序列化集成系统在使用方便性方面的优势。如何解决这一矛盾,是同时提高多模态生物特征识别系统的方便性和识别性能的关键问题。本文将半监督学习(SSL)技术引入序列化集成框架,通过提高方便性强而识别能力弱的弱特征的识别性能解决这一矛盾问题。其中的核心技术是依赖最大化降维方法(DMDR),该方法利用带标记和未标记样本及特征样本之间的对应关系,以最大化强特征与弱特征信息依赖性为目标,将弱特征投影到低维特征空间,从而实现以强特征为指导,提高弱特征识别性能的目的。在两个原型系统上,验证了半监督学习方法在提高弱特征性能方面的有效性,以及本文提出的序列化集成框架在同时提高系统使用方便性和识别性能方面的优势。

关键词： Support vector regression   neural networks   multiprocessors   message passing  

摘要： 2-dimensional Simultaneous Optical Multiprocessor Exchange Bus (2D SOME-Bus) is a reliable, robust implementation of petaflops-performance computer architecture. In this paper, we develop models to predict the performance measures (i.e. average channel utilization, average channel waiting time, average network latency, average processor utilization and average input waiting time) of a message passing architecture interconnected by the 2D SOME-Bus by using Multi-layer Feed-forward Artificial Neural Network (MFANN), Support Vector Regression (SVR) and Multiple Linear Regression (MLR). OPNET Modeler is used to simulate the message passing 2D SOME-Bus multiprocessor architecture and to create the training and testing datasets. Using 10-fold cross validation, the performance of the prediction models have been evaluated using several performance metrics. The results show that the SVR model using the radial basis function kernel (SVR-RBF) yields the lowest prediction error among all models.

关键词： 数据挖掘   决策树   不确定数据   滑坡灾害  

摘要： 中国是地质灾害的多发国家,尤其是滑坡灾害,一旦发生降雨,一些地方就很容易发生滑坡,对人们的生命造成危害,对财产造成损失。因此,如何高效的对区域滑坡灾害进行预防,是一个很重要的课题。数据挖掘是一种新兴的数据分析技术,它能通过分类方法进行学习,提取出规则,从而对未知的事物进行预测。而滑坡灾害预测受多种因素影响,其中降雨等不确定因素存在难以获取数据及有效处理等难题,为提高滑坡危险性预测的准确率,根据滑坡灾害发生相关理论及数据挖掘中决策树分类原理,本文提出了两种处理方法:基于不确定的C4.5决策树算法和不确定模糊ID3决策树算法,分别对滑坡危险性进行预测。本文首先进行了滑坡灾害和数据挖掘相关的理论介绍,为后面的章节打下理论基础。然后介绍了精度较高的不确定C4.5决策树算法,并将它应用于实例中,进行效果检验。接着在传统模糊ID3算法的基础上,利用积分思想,设计出不确定数据的模糊化处理方法,并结合模糊ID3算法,提出了一种新的不确定模糊ID3算法,建立不确定模糊ID3决策树模型,对区域滑坡危险性进行预测,在简化算法复杂度的基础上,也保证了算法的精确度。最后,将传统的算法和本文提出的算法进行比较,分析两种算法模型的优缺点,推断出各自的适用场合,为以后该类问题的算法选择提供参考。

关键词： HLA   peptide   binding   prediction   machine learning   MHC  

摘要： As major histocompatibility complexes in humans, the human leukocyte antigens (HLAs) have important functions to present antigen peptides onto T-cell receptors for immunological recognition and responses. Interpreting and predicting HLA-peptide binding are important to study T-cell epitopes, immune reactions, and the mechanisms of adverse drug reactions. We review different types of machine learning methods and tools that have been used for HLA-peptide binding prediction. We also summarize the descriptors based on which the HLA-peptide binding prediction models have been constructed and discuss the limitation and challenges of the current methods. Lastly, we give a future perspective on the HLA-peptide binding prediction method based on network analysis.

关键词： Adaptive neuro-fuzzy inference system   partial least squares regression   sequential floating forward selection   support vector regression  

摘要： The aim of this study is to propose an improved computational methodology, which is called Compressed Images for Affinity Prediction-2 (CIFAP-2) to predict binding affinities of structurally related protein-ligand complexes. CIFAP-2 method is established based on a protein-ligand model from which computational affinity information is obtained by utilizing 2D electrostatic potential images determined for the binding site of protein-ligand complexes. The quality of the prediction of the CIFAP-2 algorithm was tested using partial least squares regression (PLSR) as well as support vector regression (SVR) and adaptive neuro-fuzzy inference system (ANFIS), which are highly promising prediction methods in drug design. CIFAP-2 was applied on a protein-ligand complex system involving Caspase 3 (CASP3) and its 35 inhibitors possessing a common isatin sulfonamide pharmacophore. As a result, PLSR affinity prediction for the CASP3-ligand complexes gave rise to the most consistent information with reported empirical binding affinities (pIC50) of the CASP3 inhibitors.

关键词： RNA可变剪接   剪接位点预测   特征分析   马尔可夫模型   支持向量机  

摘要： 生物信息学的主要目标之一就是针对基因组表达与功能的研究分析,而真核生物RNA剪接机制是细胞中影响基因表达的复杂过程,剪接机制在RNA领域的研究中具有重要意义。RNA剪接机制的不同会产生不同的剪接产物,为了预测其剪接位点,就需要提出更精确的方法来解决相关的问题。机器学习是智能计算领域的重要研究点。不同与数据挖掘技术,除了对知识进行学习外,机器学习还被要求能够利用已有的知识来改善自身性能。利用机器学习方法来进行RNA剪接位点的相关研究,势必比传统的方法和工具更智能,预测结果更准确。 本文主要研究的是结合二阶马尔可夫模型与支持向量机的机器学习方法在可变剪接位点预测问题上的应用,其思想是把对剪接位点的预测转化为根据其附近序列特征进行真假位点的分类。本文主要工作有： 1、从ASD与AS-ALPS可变剪接数据库中选择数据,从中挑出五种可变剪接序列的数据集,并在剪接位点上下游各取一定长度的碱基,组建样本数据并做预处理。 2、利用基于二阶马尔可夫模型的方法进行序列特征抽取,通过分析从中选择出剪接位点二联碱基规则等主要特征,用于构成后续分类工作的特征向量。 3、利用改进了样本密集度和隶属度计算的SVM进行分类,目的是尽量减少噪声样本对预测结果准确性的负面影响。 通过测试表明,在针对可变剪接各变体位点的预测中,本文采用的预测方法比传统的算法和基本的机器学习方法的预测准确性更高,效果更好。

关键词： ANN   C&R-T   GENLIN   machine learning   shear stress distribution   small streams  

摘要： Predicting shear stress distribution has proved to be a critical problem to solve. Hence, the basic objective of this paper is to develop a prediction of shear stress distribution by machine learning algorithms including artificial neural networks, classification and regression tree, generalized linear models. The data set, which is large and feature-rich, is utilized to improve machine learning-based predictive models and extract the most important predictive factors. The 10-fold cross-validation approach was used to determine the performances of prediction methods. The predictive performances of the proposed models were found to be very close to each other. However, the results indicated that the artificial neural network, which has the R value of 0.92 +/- 0.03, achieved the best classification performance overall accuracy on the 10-fold holdout sample. The predictions of all machine learning models were well correlated with measurement data.

关键词： office buildings   knowledge   energy use   faults   Machine Learning   Building systems   office   Defect detection   Artificial Intelligence   integral structure   Energy sources   energy consumption  

摘要： To reduce energy consumption, it is important to detect energy faults, which are defined as potential sources of wasted energy in building operations. However, it is very difficult even for HVAC specialists to grasp the whole structure of energy consumption because building energy management systems (BEMSs) data are multidimensional. To address this issue, we applied machine learning approaches to analyze nonspecific energy faults using operation data in a real office building. Based on the results in this paper, we propose a diagnostic method of detecting inefficient energy use having the characteristics of both experts' empirical knowledge and a data-driven approach. We start with an existing if-then rule, compensate for the incompleteness of the knowledge with the information given by the data behavior, and attempt to extract hidden causal variables of inefficient energy use. In the proposed method, we employ three types of machine learning to widen the scope of analysis and secure adequate versatility while maintaining a high degree ofexplainability of the preliminary knowledge. The results show that this approach is potentially effective in energy faults analysis.