Kernel-Shapelets:基于卷积网络的特征子序列学习方法

在时序分类领域中,基于特征子序列的方法因能给出个体分类的局部依据,从而具有较强的可解释性,但计算十分耗时,且分类表现相较于深度学习方法并不占优。深度学习方法虽在分类表现上优于其他方法,但缺乏可解释性。为了提高基于特征子序列方法的分类表现,增强深度学习方法的可解释性,提出基于CNN网络提取特征子序列的方法Kernel-Shapelets。该方法通过提取CNN的卷积核对应权重并作进一步筛选,从而提取出特征子序列,利用全局最大池化层的输出找出CNN网络的判断依据,从而提高模型的可解释性。通过在UCR时序数据集上进行实验,Kernel-Shapelets方法的平均分类准确率为82%,相比基于特征子序列的最优基准模型提高了15.4%,证明了Kernel-Shapelets能够利用CNN网络的学习能力提取出更有效、更具有辨识性的特征子序列,提取出的特征子序列也提高了CNN网络的可解释性。

基于特征子序列修正的公交刷卡滞后时间推算

针对一票制公共交通IC卡刷卡系统存在的滞后时间问题,提出基于特征子序列修正的时差匹配推算方法。依托GPS车辆到站信息序列和刷卡聚类序列,分析首末站和中途站站间运行时间分布的差异化特征,分别从两序列中提取大时间差和连续小时间差等特征子序列;利用两种子序列间一对一或一对多的对应关系分别进行初步搜索和二次修正,兼顾匹配过程的效率和精度。将推算流程应用于山西省大同市公共汽车系统,结果显示,抽样车辆的刷卡匹配率均达到99%以上,并且针对不同类型的公共汽车线路具有较强的适应性。

周跳在高阶差分中的时序特征及精确估计

提出了独立周跳和非独立周跳的概念,给出了独立周跳在高阶差分时间序列中特征子序列,发现了非独立周跳的叠加性并给出了它在高阶差分中的特征子序列。提出以高阶差分为观测值,根据特征子序列建立观测方程,用最小二乘法探测和固定周跳的方法(探测周跳就是要判别特征子序列的位置)。对非差载波相位观测值的模拟试验表明这种方法能够精确地探测非独立周跳,而以往的高阶差分法往往无法准确探测非独立周跳。

Characteristic atom sequences of Nb-Mo alloys system in BCC structure and properties of disordered alloys

Comprehensively considering energy, volume and electronic structure of alloys, the ninth equation was determined as the interaction function of Nb-Mo alloys system in BCC structure on the basis of idea of systematic science of alloys, experimental lattice constants and heats of formation of disordered Nb（1-x）Mox alloys. The structural parameters and properties of Nb and Mo characteristic atoms sequences and corresponding characteristic crystals sequences were determined in Nb-Mo alloys system. The electronic structure and physical properties of disordered Nb（1-x）Mox alloys system were calculated according to concentration of characteristic atoms of disordered alloys. The change trend of physical properties is the same as that of electronic structure.

Cloning and Expression Patterns of a Metallothionein-like GenehtMT2 of Helianthus tuberosus

A novel cDNA sequencehtMT2, which encodes a type 2 metallothionein_like protein, was isolated from Helianthus tuberosus L. tuber cDNA library. The whole sequence is 509 bp, including an open reading frame (ORF) of 240 bp, a 5′ UTR of 62 bp and a 3′ UTR of 207 bp. Two genomic sequences covering the coding region ofhtMT2were cloned by PCR reaction. Sequence analysis revealed that the genomic sequences htMTG_1 of 986 bp and htMTG_2 of 982 bp were both composed of three exons and two introns. The deduced protein consisted of 79 amino acid residues with a predicted molecular weight of 7.8 ku (kD). Amino_terminal and carboxy_terminal domains contained 8 and 7 cysteine residues respectively, separated by a central cysteine free spacer. Sequence alignment revealed that the predicted protein ofhtMT2 was homologous to type 2 metallothioneins (MTs) of plants. Southern blotting analysis indicated that htMT2was encoded by a small multi_gene family in H. tuberosus genome. Northern blotting analysis showed that htMT2 transcripts were detected in stems, leaves and leafstalks, but no transcripts were detected in roots. The expression level in stems was the highest among the above tissues. Transcripts in stems were significantly reduced by Cu 2+ treatment. Judging from the homologies between the deduced HtMT2 and other type 2 plant metallothioneins as well as responses to metal ions, we believe thatwere cloned by PCR reaction. Sequence analysis revealed that the genomic sequences htMTG_1 of 986 bp and htMTG_2 of 982 bp were both composed of three exons and two introns. The deduced protein consisted of 79 amino acid residues with a predicted molecular weight of 7.8 ku (kD). Amino_terminal and carboxy_terminal domains contained 8 and 7 cysteine residues respectively, separated by a central cysteine free spacer. Sequence alignment revealed that the predicted protein ofhtMT2 was homologous to type 2 metallothioneins (MTs) of plants. Southern blotting analysis indicated that htMT2was encoded by a small multi_gene family in H. tuberosus genome. Northern blotting analysis showed that htMT2 transcripts were detected in stems, leaves and leafstalks, but no transcripts were detected in roots. The expression level in stems was the highest among the above tissues. Transcripts in stems were significantly reduced by Cu 2+ treatment. Judging from the homologies between the deduced HtMT2 and other type 2 plant metallothioneins as well as responses to metal ions, we believe that[ShtMT2 encodes a new type 2 metallothionein.

Product image sentence annotation based on kernel descriptors and tag-rank

Dealing with issues such as too simple image features and word noise inference in product image sentence anmotation, a product image sentence annotation model focusing on image feature learning and key words summarization is described. Three kernel descriptors such as gradient, shape, and color are extracted, respectively. Feature late-fusion is executed in turn by the multiple kernel learning model to obtain more discriminant image features. Absolute rank and relative rank of the tag-rank model are used to boost the key words＇ weights. A new word integration algorithm named word sequence blocks building （WSBB） is designed to create N-gram word sequences. Sentences are generated according to the N-gram word sequences and predefined templates. Experimental results show that both the BLEU-1 scores and BLEU-2 scores of the sentences are superior to those of the state-of-art baselines.

Molecular characterization of cyanobacterial diversity in Lake Gregory, Sri Lanka

Eutrophication or the process of nutrient enrichment of stagnant waters due to excessive use of fertilizer is becoming a critical issue worldwide. Lake Gregory, an artificial lake situated in Nuwara Eliya, Sri Lanka was once a very attractive landscape feature and recreational area attracting a large number of visitors. Rapid urbanization in surrounding areas and the consequent intensification of agricultural and industrial activities led to eutrophication and siltation in the lake. Present study was conducted to detect cyanobacterial diversity and their ability to produce hepatotoxic microcystins using polymerase chain reaction (PCR)-based techniques. Twenty five water samples (surface and bottom) were collected from the lake and total nitrogen and total carbon were estimated. Cyanobacterial cultures were grown in appropriate media and microscopic observations were used to determine the morphological diversity of cyanobacteria isolated from different sites. Genomic DNA was isolated and purified from cyanobacteria using Boom's method. DNA samples were analyzed by PCR with oligonucleotide primers for 16S rRNA gene and mcyA gene of the operon that encodes a microcystin synthetase. The 16S rRNA gene sequences revealed the presences of cyanobacteria belong to Synechococcus sp., Microcystis aeruginosa, Calothrix sp., Leptolyngbya sp., Limnothrix sp., order Oscillatoriales and order Chroococcales. The sequences obtained from this study were deposited in the database under the accession numbers (GenBank: GU368104-GU368116). PCR amplification of mcyA primers indicated the potential for toxin formation of isolated M. aeruginosa from Lake Gregory. This preliminary study shows that the Lake Gregory is under the potential risk of cyanobacterial toxicity. Clearly more work is needed to extend this finding and clarify if other cyanobacterial isolates have genetic potential to produce microcystin since this lake is utilized for recreational activities.

Panax ginseng-specific sequence characterized amplified region (SCAR) marker for testing medicinal products

To screen genetic polymorphisms of Panax ginseng, as well as those of Panax quinquefolium and Panax notoginseng, analysis of random amplified polymorphic DNA （RAPD） was performed using 120 random primers. Of the successful amplicons obtained, the Panax ginseng-specific RAPD marker C-12 was cloned into a TA vector and sequenced （Genl3ank access number KU553472）. Based on the sequence analysis results, a pair of primers specific to C-12 was designed. Finally, a SCAR marker-based identification system for Panax ginseng was developed after optimization of the reaction conditions. Using this method, two positive bands were stably observed at 300 bp and 130 bp in 33 batches of Panax ginseng samples tested, while negative results were obtained for another 101 batches of samples, including Panax quinquefolium, Panax notoginseng, adulterants, and other medicinal herbs. Thus, we successfully developed a PCR-based method for rapid and effective identification of Panax ginseng, which can be effectively used for the protection and utilization of germplasm resources and identification of the origins of Panax ginseng samples.

Fractal characteristics and its application in electromagnetic radiation signals during fracturing of coal or rock

The present study analyzed the electromagnetic radiation(EMR) time series of the destruction process of coal or rock sample under uniaxial loading and the monitoring process in working face by means of fractal geometry,and results of the correlation dimension change curve of EMR time series were obtained.In the meantime,the current study also sought the fractal characteristic to the EMR signals by contrast to the change curve of EMR signals and explored the precursory phenomenon of rock burst.This paper concluded the main findings as followed:the EMR time series of the destruction process of coal or rock sample under uniaxial loading and the monitoring process in working face corresponded to fractal;the correlation dimension of EMR time series reflected the process of coal or rock damage deformation,that is,the inner damage of coal or rock made a change from random to order.In the field application,the correlation dimension served as a new index of forecasting the coal or rock dynamic disaster.

Characteristic atom arranging crystallogphy of alloy phases for Au-Cu system

In this paper,it is pointed out that the descriptions of alloy phase structures are dependent on structural unit sequence.In the systematic science of alloys(SSA),the alloy phase structures are described by means of the symmetry element sequence combining with characteristic atom sequence.It is named the characteristic atom arranging structure,which can display the characteristic atoms at the lattice sites and the micro-inhomogeneity,besides the symmetry.Each characteristic atom has its own characters:neighboring configuration,potential energy,volume and electronic structure.The micro-inhomogeneity of alloy phases can be described by concentrations and short-range ordered parameters of characteristic atoms.The differences between the electronic structures of alloy phases and electronic structures of characteristic atoms in the alloy phases are also discussed.