基于视点差异和多分类器的三维模型分类

基于视图的3维模型分类方法与深度学习融合能有效提升模型分类的准确率。但目前的方法将相同类别的3维模型所有视点上的视图归为一类,忽略了不同视点上的视图差异,导致分类器很难学习到一个合理的分类面。为解决这一问题,该文提出一个基于深度神经网络的3维模型分类方法。该方法在3维模型的周围均匀设置多个视点组,为每个视点组训练1个视图分类器,充分挖掘不同视点组下的3维模型深度信息。这些分类器共享1个特征提取网络,但却有各自的分类网络。为了使提取的视图特征具有区分性,在特征提取网络中加入注意力机制;为了对非本视点组的视图建模,在分类网络中增加了附加类。在分类阶段首先提出一个视图选择策略,从大量视图中选择少量视图用于分类,以提高分类效率。然后提出一个分类策略通过分类视图实现可靠的3维模型分类。在ModelNet10和ModelNet40上的实验结果表明,该方法在仅用3张视图的情况下分类准确率高达93.6%和91.0%。     

一种集成式Beta过程最大间隔一类分类方法

一类分类是一种将目标类样本和其他所有的非目标类样本区分开的分类方法.传统的一类分类方法针对所有训练样本建立一个分类器,忽视了数据的内在结构,在样本分布复杂时,其分类性能会严重下降.为了提升复杂分布情况下的分类性能,该文提出一种集成式Beta过程最大间隔一类方法.该方法利用Dirichlet过程混合模型(DPM)对训练样本聚类,同时在每一个聚类学习一个Beta过程最大间隔一类分类器.通过多个分类器的集成,可以构造出一个描述能力更强的分类器,提升复杂分布下的分类效果.DPM聚类模型和Beta过程最大间隔一类分类器在同一个贝叶斯框架下联合优化,保证了每一个聚类样本的可分性.此外,在Beta过程最大间隔一类分类器中,加入了服从Beta过程先验分布的特征选择因子,从而可以降低特征冗余度以及提升分类效果.基于仿真数据、公共数据集和实测SAR图像数据的实验结果证明了所提方法的有效性.     

一种增强差异性的半监督协同分类算法

半监督学习中的Tri-Training算法打破了以往算法对充分冗余视图的限制,并通过利用三个分类器处理标记置信度和样本预测问题提高了标记效率.为进一步增强协同训练过程中分类器之间的差异性以提高性能,本文在其理论基础上提出了一种增强差异性的半监督协同分类算法.该算法利用三个不同的分类器进行学习;考虑到分类模型在更新过程中,可能会因随机抽样导致性能恶化,该算法利用基于标记类别的分层抽样法来对已标记样本集进行抽样,并通过基于分类正确率的加权投票法实现了分类器的集成,提高了预测准确率.本文通过实验对所提出算法与Tri-Training算法做了性能比较,实验结果表明本文所提出的方法在分类问题上具有较好的性能,验证了该算法的有效性和可行性.     

一种SVM集成的图像分类方法研究*

针对单分类器没有充分考虑数据集的特征而不能很好地完成分类识别,提出了一种基于集成学习技术的SVM集成的图像分类方法。该方法是在基于较为流行的词袋(Bag-of-Words,BOW)模型的图像分类方法的基础上,利用训练生成的不同SVM分类器分类测试图像,并将分类结果采用集成学习算法进行集成。分别采用传统的BOW模型的图像分类方法和本文提出的方法进行分类实验,实验结果表明采用SVM集成的图像分类方法明显提高了分类精度,具有一定的稳健性。     

多贝叶斯网络分类器集成模型研究

提出了一种多贝叶斯网络集成的分类和预测方法.把专家知识作为"疫苗",利用免疫遗传算法和约束信息熵适应度函数相结合的方法进行贝叶斯网络结构的学习,得到多个反映同一样本数据集的、网络结构复杂度折衷的、满意的贝叶斯网络结构.然后,给出了多贝叶斯网络分类器集成模型,把学习得到的贝叶斯网络进行集成,代表"专家"对未知类别的不完全数据进行群决策的分类和预测,提升贝叶斯网络分类器的泛化能力.最后,结合贝叶斯推理工具GeNIe软件,通过实例说明该方法的合理性和有效性.     

基于SVDD的半监督入侵检测研究

提出了一种基于SVDD的半监督入侵检测算法.该算法利用少量有标记正常网络数据建立两个SVDD分类器,通过相互学习来挖掘未标记数据中的隐含信息,扩大有标记正常网络数据的数量.再利用所有已标记正常网络数据用不同的单分类方法建立多个单类分类器,通过集成学习的方法得到最终的分类器.实验表明,该算法具有良好的识别性能.     

基于差异性评估对Co-training文本分类算法的改进

 Co-training算法要求两个特征视图满足一致性和独立性假设,但是,许多实际应用中不存自然的划分且满足这种假设的两个视图,且直接评估两个视图的独立性有一定的难度.分析Co-training的理论假设,本文把寻找两个满足一致性和独立性特征视图的目标,转变成寻找两个既满足一定的正确性,又存在较大的差异性的两个基分类器的问题.首先利用特征评估函数建立多个特征视图,每个特征视图包含足够的信息训练生成一个基分类器,然后通过评估基分类器之间的差异性间接评估二者的独立性,选择两个满足一定的正确性和差异性比较大的基分类器协同训练.根据每个视图上采用的分类算法是否相同,提出了两种改进算法TV-SC和TV-DC.实验表明改进的TV-SC和TV-DC算法明显优于基于随机分割特征视图的Co-Rnd算法,而且TV-DC算法的分类效果要优于TV-SC算法.     

一种集成式Beta过程最大间隔一类分类方法

一类分类是一种将目标类样本和其他所有的非目标类样本区分开的分类方法。传统的一类分类方法针对所有训练样本建立一个分类器,忽视了数据的内在结构,在样本分布复杂时,其分类性能会严重下降。为了提升复杂分布情况下的分类性能,该文提出一种集成式Beta过程最大间隔一类方法。该方法利用Dirichlet过程混合模型(DPM)对训练样本聚类,同时在每一个聚类学习一个Beta过程最大间隔一类分类器。通过多个分类器的集成,可以构造出一个描述能力更强的分类器,提升复杂分布下的分类效果。DPM聚类模型和Beta过程最大间隔一类分类器在同一个贝叶斯框架下联合优化,保证了每一个聚类样本的可分性。此外,在Beta过程最大间隔一类分类器中,加入了服从Beta过程先验分布的特征选择因子,从而可以降低特征冗余度以及提升分类效果。基于仿真数据、公共数据集和实测SAR图像数据的实验结果证明了所提方法的有效性。     

集成学习在短文本分类中的应用研究

为了进一步提高基于深度神经网络短文本分类性能,提出将集成学习方法应用于5种不同的神经网络文本分类器,即卷积神经网络、双向长短时记忆网络、卷积循环神经网络、循环卷积神经网络、分层注意力机制神经网络,分别对两种集成学习方法(Bagging,Stacking)进行了测试。实验结果表明:将多个神经网络短文本分类器进行集成的分类性能要优于单一文本分类模型;进一步两两集成的实验验证了单个模型对短文本分类性能的贡献率。     

加权解码在解决纠错输出编码Consistent-Diverse平衡问题的应用

 纠错输出编码作为解决多类分类问题的通用集成框架,能有效的把多类问题分解为二类问题从而使问题得以简化.然而在生成基分类器的过程中,经常面临提高基分类器之间的差异性和增加各基分类器与集成分类器学习的一致性的矛盾,称之为consistent-diverse平衡问题.在保证差异性的前提下减小由学习不一致性引起的分类错误率是解决该平衡问题的一个出发点,在此利用加权解码,通过对加权系数矩阵的再学习进而减弱和消除由基分类器学习不一致性产生的误差.实验利用人工数据集和UCI数据集分别加以验证,结果表明以集成分类器的分类错误率为适应度函数的遗传算法搜索出的最优加权系数矩阵相比其它方法产生的系数矩阵在解决consistent-diverse平衡问题更具有优越性.     

Bayesian Sparse Topic Model

This paper presents a new Bayesian sparse learning approach to select salient lexical features for sparse topic modeling. The Bayesian learning based on latent Dirichlet allocation (LDA) is performed by incorporating the spike-and-slab priors. According to this sparse LDA (sLDA), the spike distribution is used to select salient words while the slab distribution is applied to establish the latent topic model based on those selected relevant words. The variational inference procedure is developed to estimate prior parameters for sLDA. In the experiments on document modeling using LDA and sLDA, we find that the proposed sLDA does not only reduce the model perplexity but also reduce the memory and computation costs. Bayesian feature selection method does effectively identify relevant topic words for building sparse topic model.     

Automatic labeling of mobile apps by the type of psychological needs they satisfy

App usage is now a ubiquitous phenomenon, but little is known about what types of psychological needs are met by which apps. We proposed a method to label automatically mobile apps in terms of whether and to what extent they can satisfy users’ particular psychological needs. First, using the grounded theory approach, we conducted semi-structured in-depth interviews to identify types of needs associated with app usage. Substantive and theoretical coding of the data from the interviews as well as data from samples of app reviews yielded eight types of psychological needs app users had: utilitarian, low-cost, security, health, hedonic, social, cognitive, and self-actualization needs. Second, using the needs corpus (words and phrases) generated above, a classifier was trained using latent Dirichlet allocation (LDA) and support vector machine (SVM) algorithms to filter reviews in terms of whether they included needs-related comments. The classifier showed good performance. Finally, Labeled-LDA was used to automatically provide each review with multiple labels of the types of needs mentioned and the apps were analyzed for the different types of needs they satisfied.     

主题特征格分析:一种用户生成文本质量评估方法

本文设计了一种用户生成文本的质量分析框架.首先,基于主题分析构建商品类别主题特征集合.其次,利用主题特征与商品分类的强关联关系,构建形式化概念分析的形式背景,将分类-主题概念格化简并生成主题特征格,以此构建五个质量特征并生成质量评估模型.最后,在真实评论数据上的实验结果表明新方法具有更高预测精度.     

基于主题模型的博客标签语义知识获取

This paper focuses on semantic knowledge acquisition from blogs with the proposed tag-topic model. The model extends the Latent Dirichlet Allocation (LDA) model by adding a tag layer between the document and the topic. Each document is represented by a mixture of tags; each tag is associated with a multinomial distribution over topics and each topic is associated with a multinomial distribution over words. After parameter estimation, the tags are used to describe the underlying topics. Thus the latent semantic knowledge within the topics could be represented explicitly. The tags are treated as concepts, and the top-N words from the top topics are selected as related words of the concepts. Then PMI-IR is employed to compute the relatedness between each tag-word pair and noisy words with low correlation removed to improve the quality of the semantic knowledge. Experiment results show that the proposed method can effectively capture semantic knowledge, especially the polyseme and synonym.     

Multi-view representation learning for multi-view action recognition

Although multiple methods have been proposed for human action recognition, the existing multi-view approaches cannot well discover meaningful relationship among multiple action categories from different views. To handle this problem, this paper proposes an multi-view learning approach for multi-view action recognition. First, the proposed method leverages the popular visual representation method, bag-of-visual-words (BoVW)/fisher vector (FV), to represent individual videos in each view. Second, the sparse coding algorithm is utilized to transfer the low-level features of various views into the discriminative and high-level semantics space. Third, we employ the multi-task learning (MTL) approach for joint action modeling and discovery of latent relationship among different action categories. The extensive experimental results on M²I and IXMAS datasets have demonstrated the effectiveness of our proposed approach. Moreover, the experiments further demonstrate that the discovered latent relationship can benefit multi-view model learning to augment the performance of action recognition.     

Multitier ensemble classifiers for malicious network traffic detection

A malicious network traffic detection method based on multi-level distributed ensemble classifier was proposed for the problem that the attack model was not trained accurately due to the lack of some samples of attack steps for detecting attack in the current network big data environment,as well as the deficiency of the existing ensemble classifier in the construction of multilevel classifier.The dataset was first preprocessed and aggregated into different clusters,then noise processing on each cluster was performed,and then a multi-level distributed ensemble classifier,MLDE,was built to detect network malicious traffic.In the MLDE ensemble framework the base classifier was used at the bottom,while the non-bottom different ensemble classifiers were used.The framework was simple to be built.In the framework,big data sets were concurrently processed,and the size of ensemble classifier was adjusted according to the size of data sets.The experimental results show that the AUC value can reach 0.999 when MLDE base users random forest was used in the first layer,bagging was used in the second layer and AdaBoost classifier was used in the third layer.     

Improvement of LDA model with time factor for collaborative filtering

Collaborative filtering (CF) is one of the most widely used Algorithm in recommender systems, which help users obtain the information they may like. We proposed a latent Dirichlet allocation (LDA) model combining time and rating (TR-LDA) for CF. We use mathematical methods to fit the Ebbinghaus forgetting curve in our method and introduce time weights based on time window to find out the impact of time on user's interests. The user's choice of items is not only influenced by his/ her interests, but also influenced by other's rating. According to the users' feedback, we find their rating distribution on items under the interests. Finally, experimental results on real data sets MovieLens 100 k and MovieLens 1 M show that the proposed Algorithm can predict the user implicit interests effectively and improve the recommendation performance.     

一种面向隐含主题的上下文树核

该文针对上下文树核用于文本表示时缺乏语义信息的问题,提出了一种面向隐含主题的上下文树核构造方法。首先采用隐含狄利克雷分配将文本中的词语映射到隐含主题空间,然后以隐含主题为单位建立上下文树模型,最后利用模型间的互信息构造上下文树核。该方法以词的语义类别来定义文本的生成模型,解决了基于词的文本建模时所遇到的统计数据的稀疏性问题。在文本数据集上的聚类实验结果表明,文中提出的上下文树核能够更好地度量文本间主题的相似性,提高了文本聚类的性能。     

融合链接拓扑结构和用户兴趣的朋友推荐方法

提出一种新的朋友推荐方法,该方法同时使用用户兴趣和朋友关系这2种因素来为目标用户推荐朋友,对PageRank算法进行改进,提出一种能同时融合上述2种因素的Topic_Friend_PageRank(TFPR)模型。首先,采用LDA(latent Dirichlet allocation)分析用户发布的消息内容,将用户表示为若干主题上的分布,从而建模用户的兴趣。接下来,使用加权的PageRank算法建模用户在整个链接拓扑中的重要程度和用户之间朋友关系的相似性。最后根据主题感知的PageRank思想,将用户兴趣融入前面提到的加权PageRank中,形成同时融合用户兴趣和朋友关系的TFPR模型。采用新浪微博数据验证所提模型的性能,实验证明该模型能同时得到较高的准确率和召回率。