Linear dimensionality reduction: Survey, insights, and generalizations

© 2015 John P. Cunningham and Zoubin Ghahramani. Linear dimensionality reduction methods are a cornerstone of analyzing high dimensional data, due to their simple geometric interpretations and typically attractive computational properties. These methods capture many data features of interest, such as covariance, dynamical structure, correlation between data sets, input-output relationships, and margin between data classes. Methods have been developed with a variety of names and motivations in many fields, and perhaps as a result the connections between all these methods have not been highlighted. Here we survey methods from this disparate literature as optimization programs over matrix manifolds. We discuss principal component analysis, factor analysis, linear multidimensional scaling, Fisher's linear discriminant analysis, canonical correlations analysis, maximum autocorrelation factors, slow feature analysis, sufficient dimensionality reduction, undercomplete independent component analysis, linear regression, distance metric learning, and more. This optimization framework gives insight to some rarely discussed shortcomings of well-known methods, such as the suboptimality of certain eigenvector solutions. Modern techniques for optimization over matrix manifolds enable a generic linear dimensionality reduction solver, which accepts as input data and an objective to be optimized, and returns, as output, an optimal low-dimensional projection of the data. This simple optimization framework further allows straightforward generalizations and novel variants of classical methods, which we demonstrate here by creating an orthogonal-projection canonical correlations analysis. More broadly, this survey and generic solver suggest that linear dimensionality reduction can move toward becoming a blackbox, objective-agnostic numerical technology.

Identification of a novel C1q family member in color crucian carp (Carassius auratus) ovary

Potential roles of Clq/tumor necrosis factor (TNF) superfamily proteins have been observed in vertebrate oogenesis and oocyte maturation, but no ovary-specific member has been identified so far. In this study, we have cloned and identified a novel member of Clq family with a Clq domain in the C-terminal from fully grown oocyte cDNA library of color crucian carp and demonstrated that the gene might be specifically expressed in ovary and therefore designated as Carassius auratus ovary-specific Clq-like factor, CaOClq-like factor. It encodes a 213 amino acid protein with a 17 amino acid signal peptide. There is only one protein band of about 24.5 kDa in the extracts from phase I to phase IV oocytes, but two positive protein bands are detected in the extracts of mature eggs and fertilized eggs. Furthermore, the mobility shift of the smaller target protein band cannot be eliminated by phosphatase treatment, but the larger protein band increases its mobility on the gel after phosphatase treatment, suggesting that the larger protein might be a phosphorylated form. Immunofluorescence localization indicates that the CaOClq-like proteins localize in cytoplasm, cytoplasm membrane and egg envelope of the oocytes at cortical granule stage and vitellogenesis stage, whereas they were compressed to cytoplasm margin in ovulated mature eggs and discharged into perivitelline space between cytoplasm membrane and egg envelope after egg fertilization. Further studies on distribution and translocation mechanism of the CaOClq-like factor will be benefit to elucidate the unique function in oogenesis, oocyte maturation and egg fertilization. (C) 2004 Elsevier Inc. All rights reserved.

Revision of the cyprinid genus Placocheilus Wu, 1977 in China, with description of a new species from Yunnan

Placocheilus is re-diagnosed on the basis of oromandibular structures compared to those of closely related groups. It is represented in China by three species restricted to Yunnan, namely P. caudofasciatus from the Tengtiaohe River (Nam Na), P. cryptonemus from the Nujiang Rive ( upper Salween River basin), and P. robustus, herein described as a new species from the Yuanjiang River ( Red River basin). P. robustus n. sp. is differentiated from its congeners in having a stout caudal peduncle ( depth 72.4-82.5% of its length) and the medium-sized scaleless midventral region of belly extended slightly beyond halfway from the pectoral- to pelvic-fin origin. It is further distinguished from P. caudofasciatus in having a coarsely crenulated rostral fold with 18-23 indentations on its distal margin and a smaller mental adhesive disc ( length 66.7-71.4% of its width and 37.0-43.1% of the head length).

Notholca dongtingensis (Rotifera : Monogononta : Brachionidae), a new species from Dongting Lake, China

Notholca dongtingensis n.sp. was found in the second largest lake of China, Dongting Lake. It is related to Notholca labis Gosse, 1887 and N. kozhovi Vassilijewa & Kutikova, 1969. Its main distinguishing taxonomic features are a square-oval lorica, very short anterior spines, anterior lateral spines curving outwards and protrusile posterior margin of the ventral plate.

Biologically Inspired Tensor Features

According to the research results reported in the past decades, it is well acknowledged that face recognition is not a trivial task. With the development of electronic devices, we are gradually revealing the secret of object recognition in the primate's visual cortex. Therefore, it is time to reconsider face recognition by using biologically inspired features. In this paper, we represent face images by utilizing the C1 units, which correspond to complex cells in the visual cortex, and pool over S1 units by using a maximum operation to reserve only the maximum response of each local area of S1 units. The new representation is termed C1 Face. Because C1 Face is naturally a third-order tensor (or a three dimensional array), we propose three-way discriminative locality alignment (TWDLA), an extension of the discriminative locality alignment, which is a top-level discriminate manifold learning-based subspace learning algorithm. TWDLA has the following advantages: (1) it takes third-order tensors as input directly so the structure information can be well preserved; (2) it models the local geometry over every modality of the input tensors so the spatial relations of input tensors within a class can be preserved; (3) it maximizes the margin between a tensor and tensors from other classes over each modality so it performs well for recognition tasks and (4) it has no under sampling problem. Extensive experiments on YALE and FERET datasets show (1) the proposed C1Face representation can better represent face images than raw pixels and (2) TWDLA can duly preserve both the local geometry and the discriminative information over every modality for recognition.

近邻边界Fisher判别分析

将数据集进行合理的维数约简对于一些机器学习算法效率的提高起着至关重要的影响.该文提出了一种利用数据点邻域信息的线性监督降维算法:近邻边界Fisher判别分析(Neighborhood Margin Fisher Discriminant Analysis,NMFDA).NMFDA尝试将每一数据点邻域内最远的同类数据点和最近的异类数据点之间的边界在投影子空间内尽可能地扩大,从而提高基于距离的识别算法的准确率.同时为了解决非线性降维问题,提出了Kernel NMFDA,通过在几个标准人脸数据库上与其它降维算法的对比识别实验,验证了提出算法的有效性.

A 10-bit 50-MS/s sample-and-hold circuit with low distortion sampling switches

A fully-differential switched-capacitor sample-and-hold (S/H) circuit used in a 10-bit 50-MS/s pipeline analog-to-digital converter (ADC) was designed and fabricated using a 0.35-μm CMOS process. Capacitor fliparound architecture was used in the S/H circuit to lower the power consumption. In addition, a gain-boosted operational transconductance amplifier (OTA) was designed with a DC gain of 94 dB and a unit gain bandwidth of 460 MHz at a phase margin of 63 degree, which matches the S/H circuit. A novel double-side bootstrapped switch was used, improving the precision of the whole circuit. The measured results have shown that the S/H circuit reaches a spurious free dynamic range (SFDR) of 67 dB and a signal-to-noise ratio (SNR) of 62.1 dB for a 2.5 MHz input signal with 50 MS/s sampling rate. The 0.12 mm~2 S/H circuit operates from a 3.3 V supply and consumes 13.6 mW.

A Novel 4T nMOS-Only SRAM Cell in 32nm Technology Node

This paper proposes a novel loadless 4T SRAM cell composed of nMOS transistors. The SRAM cell is based on 32nm silicon-on-insulator （SO1） technology node. It consists of two access transistors and two pull-down transistors. The pull-down transistors have larger channel length than the access transistors. Due to the significant short channel effect of small-size MOS transistors, the access transistors have much larger leakage current than the pull-down transistors,enabling the SRAM cell to maintain logic ＂1＂ while in standby. The storage node voltages of the cell are fed back to the back-gates of the access transistors,enabling the stable ＂read＂ operation of the cell. The use of back-gate feedback also helps to im- prove the static noise margin （SNM） of the cell. The proposed SRAM cell has smaller area than conventional bulk 6T SRAM cells and 4T SRAM cells. The speed and power dissipation of the SRAM cell are simulated and discussed. The SRAM cell can operate with a 0. 5V supply voltage.

青藏高原东缘川西云杉林皆伐后灌木生长、繁殖与更新的研究

植物群落及其环境在干扰后的演替格局和过程的研究，是群落和生态系统动态研究的一个热点。选取青藏高原东缘山原区川西云杉林皆伐后，从草地过渡到灌丛的关键阶段的4 个皆伐迹地（恢复时间为8 a、10 a、16 a 和21 a），研究皆伐及自然恢复过程对林下典型灌木银露梅（Potentilla glabra）和唐古特忍冬（Lonicera tangutica）的生长与繁殖能力的影响以及灌木植物在迹地上的更新情况，分析灌木在不同生境中的适应对策和适应能力的差异,为揭示青藏高原东缘山原区迹地植被从草甸到灌丛演替的过程和特点及促进迹地演替与植被恢复进程提供理论依据和技术支撑。研究主要结论如下：1）皆伐后银露梅生长和繁殖能力显著提高，但对唐古特忍冬的影响不明显。皆伐后银露梅丛基径、高度和各部分生物量都显著增加（P < 0.05），但唐古特忍冬只有叶生物量和地下生物量增加，总生物量和其余构件生物量无显著变化。皆伐后，银露梅的结实数量、结实株数、不结实株数和结实株/不结实株比例显著增加（P < 0.05）。自然恢复过程中，银露梅和唐古特忍冬生长能力以及银露梅的结实量都表现出降低的趋势。随着迹地自然恢复时间的增加，银露梅和唐古特忍冬的基径、高度、丛叶片数和各部分生物量有减少的趋势。银露梅的结实株数增加, 但结实数量减少。2）皆伐对银露梅和唐古特忍冬生物量分配模式影响不一致。原始林和迹地中（除CT85）银露梅的生物量大小关系皆为：地下>茎>侧枝>叶。唐古特忍冬在原始林中的生物量大小关系为：茎＞地下部分＞侧枝＞叶,而皆伐后生物量的分配情况改变，生物量大小关系变为：地下部分＞茎＞侧枝＞叶。随着自然恢复时间的增加，银露梅减少了地下生物量的分配，而唐古特忍冬增加了地下生物量的分配。3）皆伐和恢复时间的增加改变了迹地物种组成，促使阳性乔、灌木在迹地上定居。4 个迹地上共出现了灌木15 种，乔木3 种，没有出现天然云杉和冷杉幼苗。随着恢复时间的增加，迹地上的灌木物种由原始林下的耐阴物种逐步发展为以针刺悬钩子（Rubus pungens）为主的阳性灌木。4）影响灌木幼苗密度和幼树密度的因子不一致。灌木幼苗密度与灌木层盖度显著负相关，与苔藓层盖度显著正相关。幼树密度与草本层盖度正相关，与苔藓层盖度、灌木层盖度和高度负相关。5）研究发现在青藏高原东缘山原区皆伐15～20 a 后，迹地仍以草本植物为主，推测皆伐后至少20 a 以上迹地才可能向灌丛阶段过渡，比高山峡谷地区的演替进程至少推迟了20 a。银露梅和唐古特忍冬在皆伐后自然恢复过程中表现出不同的生长与繁殖策略是由两个物种的生物学特性的差异引起的。银露梅比唐古特忍冬更适应迹地退化环境。促进青藏高原东缘山原林区迹地森林恢复一方面是尽量减少人为活动的破坏，另一方面，可以通过在迹地中播种适当的乡土乔、灌木种子（如白桦、银露梅）等人工措施，以加快演替进程。The succession pattern and process of plant community and their environments is a hot spotin community and ecosystem dynamic study. Four clearcuts were chosen in Rangtang(recovery time of 8 a、10 a、16 a and 21 a), which represented the key stage of thecommunity evolved from grass stage to shrub stage in the eastern margin of theQinghai-Tibetan Plateau. The growth and reproduction of the Potentilla glabra andLonicera tangutica and the natural regeneration of shrub plants in the primary Piceabalfouriana forest and 4 clearcuts were studied to explore how clear cutting andnatural recovery process affected the understory shrub species during the 21 years inthe eastern margin of the Qinghai-Tibetan Plateau. The main results were below.1) The growth and reproduction of P. glabra significantly increased after forestclear cutting.. But it was not so significant as to the L. tangutica. The organismbiomass and total biomass of P. glabra were increased obviously after clear cutting（P< 0.05）. But only leaves and underground biomass of L. tangutica increasedsignificantly after clear cutting（P < 0.05）. The number of fruit and growth of P. glabraincreased significantly after clear cutting too（P < 0.05）. The ramet height, basaldiameter , organism biomass and friut number of P. glabra and L. tangutica reducedas the increase of recovery time.2) The biomass allocation patterns varied between P. glabra and L. tangutica inthe primary forest and clearcuts. The biomass allocation of P. glabra both in primary forest and clearcuts was followed as: underground part > stem > branch > leaves.However, the biomass allocation of L. tangutica had changed after the clear cutting.The biomass allocation of L. tangutica in the primary forest was followed as: stem >underground part > branch > leaves and it was underground part > stem > branch >leave in clearcuts. The biomass allocation of P. glabra and L. tangutica varied amongclearcuts. Aboveground biomass was increased while underground biomass decreasedfor P. glabra with the increase of recovery period. However, the L. tangutica showedthe reverse changing pattern.3) Clear cutting and recovery time had changed the species composition of theclearcuts. There were 15.shruby species and 3 tree species in the four clearcuts. Nospruce and fir seedlings were found. In the early stage after clear cutting, there wereonly understory shrub species from the primary Picea balfouriana forest. The sunnyspecies, especieally Rubus pungens invaded intensly as the increase of recovery time.4) There was a significant negative relationship between density of seedlingswith shrub layer coverage and positive correlation with moss coverage. The saplingshad significantly positive correlation with herb layer coverage and negativecorrelation with moss coverage, shrub layer coverage and height.5）Comparing to studies in Miyalou, a nearby high mountain and canyon area,the secondary sucession in this subalpine plateau areas lagged at least 20 years.P. glabra and L. tangutica showed different growth and reproduction strategies toclear cutting and natural recovery , which may associated with the difference of theirbiological characters. P. glabra was more adaptive to the clear cutting than the L.tangutica. Two suggestions were probably recommended to promote the recoveryprogress in the subalpine plateau areas based on the results of this study. Limitanthropogenic disturbance, and meanwhile sow native tree and shrub seeds inclearcuts.

Tectonic geomorphology of the Ryukyu Trench-Arc-Backarc System: geological-geophysical exploration and mapping

OKINAWA TROUGH; BASIN

Characterization of proteolytic bacteria from the Aleutian deep-sea and their proteases

Six deep-sea proteolytic bacteria taken from Aleutian margin sediments were screened; one of them produced a cold-adapted neutral halophilic protease. These bacteria belong to Pseudoalteromonas spp., which were identified by the 16S rDNA sequence. Of the six proteases produced, two were neutral cold-adapted proteases that showed their optimal activity at pH 7-8 and at temperature close to 35 degrees C, and the other four were alkaline proteases that showed their optimal activity at pH 9 and at temperature of 40-45 degrees C. The neutral cold-adapted protease E1 showed its optimal activity at a sodium chloride concentration of 2 M, whereas the activity of the other five proteases decreased at elevated sodium chloride concentrations. Protease E1 was purified to electrophoretic homogeneity and its molecular mass was 34 kDa, as estimated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). The molecular weight of protease E1 was determined to be 32,411 Da by mass spectrometric analysis. Phenylmethyl sulfonylfluoride (PMSF) did not inhibit the activity of this protease, whereas it was partially inhibited by ethylenediaminetetra-acetic acid sodium salt (EDTA-Na). De novo amino acid sequencing proved protease E1 to be a novel protein.

Diversity and Distribution of Sediment NirS-Encoding Bacterial Assemblages in Response to Environmental Gradients in the Eutrophied Jiaozhou Bay, China

A gene-clone-library-based molecular approach was used to study the nirS-encoding bacteria-environment relationship in the sediments of the eutrophic Jiaozhou Bay. Diverse nirS sequences were recovered and most of them were related to the marine cluster I group, ubiquitous in estuarine, coastal, and marine environments. Some NirS sequences were unique to the Jiaozhou Bay, such as the marine subcluster VIIg sequences. Most of the Jiaozhou Bay NirS sequences had their closest matches originally detected in estuarine and marine sediments, especially from the Chesapeake Bay, indicating similarity of the denitrifying bacterial communities in similar coastal environments in spite of geographical distance. Multivariate statistical analyses indicated that the spatial distribution of the nirS-encoding bacterial assemblages is highly correlated with environmental factors, such as sediment silt content, NH4+ concentration, and OrgC/OrgN. The nirS-encoding bacterial assemblages in the most hypernutrified stations could be easily distinguished from that of the least eutrophic station. For the first time, the sedimentological condition was found to influence the structure and distribution of the sediment denitrifying bacterial community.

Tectonic Evolution and Dynamics of Deepwater Area of Pearl River Mouth Basin, Northern South China Sea

Quantitative studies on the evolution and dynamics of the deepwater area of Pearl River Mouth basin (PRMB) were carried out based on the latest geological and seismic data. The study area is generally in an extensional state during the Cenozoic. The major extension happened in the earlier syn-rift stages before 23 Ma and the extension after 23 Ma is negligible. Two rapid subsidence periods, 32-23 Ma and 5.3-2.6 Ma, are identified, which are related to the abrupt heat decay during margin breakup and the collision between the Philippine Sea plate and the Eurasian plate, respectively. The strongest crustal thinning in the Baiyun (sic) sag may trigger the syn-rift volcanism along the weak faulted belt around the sag. The Cenozoic tectonic evolution of the study area could be divided into five stages: rifting (similar to 50-40 Ma), rift-drift transition (similar to 40-32 Ma), early post-breakup (similar to 32-23 Ma), thermal subsidence (similar to 23-5.3 Ma) and neotectonic movement (similar to 5.3-0 Ma).

Genetic model of the hydrate system in the fine grain sediments in the northern continental slope of South China Sea

Gas hydrate samples were obtained firstly in China by drilling on the northern margin of South China Sea (SCS). To understand the formation mechanism of this unique accumulation system, this paper discusses the factors controlling the formation of the system by accurate geophysical interpretation and geological analysis, based on the high precision 2-D and 3-D multichannel seismic data in the drilling area. There are three key factors controlling the accumulation of the gas hydrate system in fine grain sediment: (1) large volume of fluid bearing methane gas Joins the formation of gas hydrate. Active fluid flow in the northern South China Sea makes both thermal gas and/or biogenic gas migrate into shallow strata and form hydrate in the gas hydrate stability zone (GHSZ). The fluid flow includes mud diapir and gas chimney structure. They are commonly characterized by positive topographic relief, acoustic turbidity and push-down, and low reflection intensity on seismic profiles. The gas chimneys can reach to GHSZ, which favors the development of BSRs. It means that the active fluid flow has a close relationship with the formation and accumulation of gas hydrate. (2) The episodic process of fracture plays an important role in the generation of gas hydrate. It may provide the passage along which thermogenic or biogenic gas migrated into gas hydrate stability zone (GHSZ) upward. And it increases the pore space for the growth of hydrate crystal. (3) Submarine landslide induced the anomalous overpressure activity and development of fracture in the GHSZ. The formation model of high concentration gas hydrate in the drilling sea area was proposed on the basis of above analysis.