Stable local dimensionality reduction approaches

Dimensionality reduction is a big challenge in many areas. A large number of local approaches, stemming from statistics or geometry, have been developed. However, in practice these local approaches are often in lack of robustness, since in contrast to maximum variance unfolding (MVU), which explicitly unfolds the manifold, they merely characterize local geometry structure. Moreover, the eigenproblems that they encounter, are hard to solve. We propose a unified framework that explicitly unfolds the manifold and reformulate local approaches as the semi-definite programs instead of the above-mentioned eigenproblems. Three well-known algorithms, locally linear embedding (LLE), laplacian eigenmaps (LE) and local tangent space alignment (LTSA) are reinterpreted and improved within this framework. Several experiments are presented to demonstrate the potential of our framework and the improvements of these local algorithms.     

Embedding fault-free cycles in crossed cubes with conditional link faults

The crossed cube, which is a variation of the hypercube, possesses some properties that are superior to those of the hypercube. In this paper, we show that with the assumption of each node incident with at least two fault-free links, an n-dimensional crossed cube with up to 2n−5 link faults can embed, with dilation one, fault-free cycles of lengths ranging from 4 to 2 ⁿ . The assumption is meaningful, for its occurrence probability is very close to 1, and the result is optimal with respect to the number of link faults tolerated. Consequently, it is very probable that algorithms executable on rings of lengths ranging from 4 to 2 ⁿ can be applied to an n-dimensional crossed cube with up to 2n−5 link faults.

一种基于自适应量化嵌入器的隐写方法

感知质量和容量一直是图像隐写算法设计中难以折中的两个要素,如何取得更好的综合性能已吸引了很多研究者的兴趣.提出一种基于自适应量化嵌入器的隐写方法.这一方法通过改变某一载体信号所在子空间的索引来调制秘密信息,自适应特性体现在决定着子空间尺寸的分割码本的设计上.将这种隐写方法用于以2×2图像块为单元的数字图像隐写算法设计中,并将分割码本设计得具备HVS特性以取得较好的折中效果.实验表明,算法在感知质量和容量方面具有较好的综合性能.     

有监督核化邻域投影分析算法

通过将鉴别邻域嵌入分析算法扩展到非线性场景,提出了一种有监督核化邻域投影分析算法。该算法在目标函数中引入类别标签和线性投影矩阵,并利用核函数处理非线性数据。通过两种不同策略优化目标函数,可将该算法进一步细分为有监督核化邻域投影分析算法一及有监督核化邻域投影分析算法二。其中,在有监督核化邻域投影分析算法一中应用拉普拉斯搜索方向达到了较快的收敛速度并降低了计算复杂度。实验结果表明,所提算法对于复杂的数据流形具有较高的识别率,且与鉴别邻域嵌入分析等相关算法相比在有效性和鲁棒性方面的表现更为出色。     

深度学习中汉语字向量和词向量结合方式探究

该文旨在探究深度学习中汉语字向量和词向量的有效结合方式。我们在以词作为基础语义单元和以字作为基础语义单元这两个方向进行探究,实验了字、词信息多种浅层结合方式和深层结合方式。为了验证该文提出的结合方式的有效性,我们改进了一种compare-aggregate模型,并在基于文档的问答系统上进行了实验。实验结果表明,有效的汉语字向量和词向量的结合方式超越了单独的字向量和词向量,提升了基于文档的问答系统的性能,使其结果与目前最好的结果可媲美。     

基于情感词向量的微博情感分类

该文提出了一种基于情感词向量的情感分类方法。词向量采用连续实数域上的固定维数向量来表示词汇,能够表达词汇丰富的语义信息。词向量的学习方法,如word2vec,能从大规模语料中通过上下文信息挖掘出潜藏的词语间语义关联。本文在从语料中学习得到的蕴含语义信息的词向量基础上,对其进行情感调整,得到同时考虑语义和情感倾向的词向量。对于一篇输入文本,基于情感词向量建立文本的特征表示,采用机器学习的方法对文本进行情感分类。该方法与基于词、N-gram及原始word2vec词向量构建文本表示的方法相比,情感分类准确率更高、性能和稳定性更好。     

基于信息熵与动态区域分割的半脆弱数字水印算法

已有的半脆弱水印算法大多数采取了双步长的固定量化方法,对载体图像没有针对攻击特性给予考虑,只将载体图像分为平滑和纹理两个区域,使得水印的鲁棒性进入了一个瓶颈状态。为了进一步提高水印的鲁棒性,提出了一种基于图像动态分割技术和信息熵的半脆弱水印算法。图像动态分割技术将图像分为若干个嵌入区域,并且确定各个嵌入区域的嵌入强度;基于信息熵的步长量化算法将信息熵引入到步长量化的计算方法中,能够有效衡量不同嵌入区域所携带的敏感信息量。实验结果表明,提出的算法与已有半脆弱水印算法相比,视觉掩蔽性好,鲁棒性也有了较明显的提高。     

A new neural network algorithm for planarization problems

To deal with the planarization problem widely used in many applications including routing very-large-scale integration （VLSI） circuits, this paper points out that only when its vertices are arranged in some specific order in a line can a planar graph be embedded on a line without any cross connections or cross edges. Energy function is proposed to meet the need of embedding a graph on a single line and route it correctly. A Hopfield network is designed according to the proposed energy function for such embedding and routing. The advantage of the proposed method is that it not only can detect if a graph is a planar one or not, but also can embed a planar graph or the maximal planar subgraph of a non-planar graph on a single line. In addition, simulated annealing is employed for helping the network to escape from local minima during the running of the Hopfield network. Experiments of the proposed method and its comparison with some existent conventional methods were performed and the results indicate that the proposed method is of great feasibility and effectiveness especially for the planarization problem of large graphs.