Local Structure Of Samples Research Articles

Deep metric learning assisted by intra-variance in a semi-supervised view of learning

Deep metric learning aims to construct an embedding space where samples belonging to the same class are closely grouped together, while samples from different classes are widely separated. Many existing deep metric learning methods focus on maximizing the difference between inter-class features, and semantic related information is obtained by increasing the distance between samples of different classes in the embedding space. However, by compressing all positive samples together and creating large margins between different classes, the local structure between similar samples is inadvertently disrupted. This disregard for the intra-class variance presents in the local structure results in an embedding space that exhibits lower generalizability when faced with unseen classes. Consequently, the network tends to overfit the training set and performs poorly on the test set, potentially leading to crashes during evaluation. Taking this into account, this paper introduces a self-supervised generative assisted ranking framework that offers a semi-supervised perspective on learning intra-class variance within traditional supervised deep metric learning. Specifically, this paper employs various intensities and diversities to synthesize samples based on positive examples, aiming to simulate the complex variations between similar samples. Then, an intra-class ranking loss function is devised based on the principles of self-supervised learning. This loss function constrains the ordering relationship of synthesized samples according to their generation intensity, enabling the network to capture subtle intra-class differences and maintain the intra-class distribution. With the adoption of this approach, a more realistic embedding space can be achieved, preserving both the global and local structures of samples.

A Cross-Media Advertising Design and Communication Model Based on Feature Subspace Learning.

This paper uses feature subspace learning and cross-media retrieval analysis to construct an advertising design and communication model. To address the problems of the traditional feature subspace learning model and make the samples effectively maintain their local structure and discriminative properties after projection into the feature space, this paper proposes a discriminative feature subspace learning model based on Low-Rank Representation (LRR), which explores the local structure of samples through Low-Rank Representation and uses the representation coefficients as similarity constraints of samples in the projection space so that the projection subspace can better maintain the local nearest-neighbor relationship of samples. Based on the common subspace learning, this paper uses the extreme learning machine method to improve the cross-modal retrieval accuracy, mining deeper data features and maximizing the correlation between different modalities, so that the learned shared subspace is more discriminative; meanwhile, it proposes realizing cross-modal retrieval by the deep convolutional generative adversarial network, using unlabeled samples to further explore the correlation of different modal data and improve the cross-modal performance. The clustering quality of images and audios is corrected in the feature subspace obtained by dimensionality reduction through an optimization algorithm based on similarity transfer. Three active learning strategies are designed to calculate the conditional probability of unannotated samples around user-annotated samples in the correlation feedback process, thus improving the efficiency of cross-media retrieval in the case of limited feedback samples. The experimental results show that the method accurately measures the cross-media relevance and effectively achieves mutual retrieval between image and audio data. Through the study of cross-media advertising design and communication models based on feature subspace learning, it is of positive significance to advance commercial advertising design by guiding designers and artists to better utilize digital media technology for artistic design activities at the level of theoretical research and applied practice.

Study of carbon doped cobalt mononitride thin films

In this work, we address the issue of N diffusion in cobalt mononitride (CoN) thin films. The diffusivity of N in CoN is the highest among 3d-transition metal nitrides, causing a gradient in N concentration across the depth. In order to reduce N-diffusion and to improve the thermal stability, we doped various amounts of C in CoN. The long-range ordering of samples was investigated using x-ray diffraction. Elemental depth profiles were investigated using secondary ion mass spectroscopy, exhibiting a gradient in the depth profile towards the surface of undoped CoN film and the gradient decreases with an increase in C-doping. The local structure of samples was studied using x-ray absorption near-edge spectroscopy at N and Co K-edges and extended x-ray absorption fine structure (EXAFS) at Co K-edge. Defects and vacancies due to N-diffusion can also be seen from the EXAFS spectra of the undoped sample which minimizes at some optimum amount of C-doping. The thermal stability also gets improved at this optimum amount of C-doping. The obtained results can be utilized in the energy sector in which CoN finds many applications such as a high capacity electrode for Li-ion batteries, non-aqueous supercapacitors, and solar cells.

Synthesis and structural investigation of stoichiometric iron mononitride thin films

We report the synthesis of iron mononitride (FeN) thin films having zinc blende (ZB) type structure, deposited using a direct current magnetron sputtering process on amorphous quartz substrates. The effect of substrate temperature (Ts) on the structure of FeN thin films was systematically investigated by varying the Tsfrom 300 to 500 K(in close intervals of 25K). The long-range as well as the local structure of samples was studied using x-ray diffraction (XRD), x-ray absorption near edge spectroscopy (XANES) at N K-edge, Fe L and K-edge and extended x-ray absorption fine structure (EXAFS) at Fe K-edge. The lattice parameter (LP) obtained from XRD and the presence of an intense pre-edge feature in Fe K-edge XANES confirms the ZB-structure of FeN samples. It was found in this work that the stoichiometric FeN with a LP close to its theoretical value can only be obtained in a narrow temperature regime due to minimization of stresses generated during the growth near the thermodynamic equilibrium conditions. These findings were further validated by EXAFS results yielding Fe–N and Fe–Fe bond distances, in conformity with the theoretically predicted values for FeN in ZB type structure.

Multi-bit quantization based on neighboring structure preservation

Hashing based approximate nearest neighbor search has become a research hotspot in computer vision. Most existing hashing methods concentrate on projection learning, and few efforts are dedicated to quantization coding. In this paper, we present a multi-bit quantization strategy to improve the quantization quality of projection values by adaptively learning quantization thresholds and quantizing each projection dimension with multiple bits. Our method exploits both the similarity and the local structure of samples in the original feature space and the pair-wise samples coding consistency. Extensive experiments on two canonical image datasets have shown that our method consistently outperforms the state-of-the-art quantization methods in terms of query performance.

Study of Co-Cr Mixed Oxides for Different Applications

Mixture metal oxide system Co-Cr were prepared by decomposition of the precursor complexes obtained from a mixture Co-Cr nitrates with tartaric acid. The samples were subsequently submitted to chemical analysis, magnetic measurements, IR spectrometry, UV-Vis, thermal analysis, XRD, EXAFS and TPR. At 600oC was identificated a solid solution of CoCr2O4 with Co3O4. With Cu-CrK-edge EXAFS the local structure of samples calcinated of 420oC and 600oC was obtained. By theirs analysis was calculated: the lengths of M-O bounds, number of coordination (N), Debye-Waller factor (A1) and x metallic cations fraction in tetrahedral structures (spinelic centers). The TPR measurements were emphasized the existence of two reduction cycles.

Semi-Supervised Classification Based on Mixture Graph

Graph-based semi-supervised classification heavily depends on a well-structured graph. In this paper, we investigate a mixture graph and propose a method called semi-supervised classification based on mixture graph (SSCMG). SSCMG first constructs multiple k nearest neighborhood (kNN) graphs in different random subspaces of the samples. Then, it combines these graphs into a mixture graph and incorporates this graph into a graph-based semi-supervised classifier. SSCMG can preserve the local structure of samples in subspaces and is less affected by noisy and redundant features. Empirical study on facial images classification shows that SSCMG not only has better recognition performance, but also is more robust to input parameters than other related methods.

X-ray nano-diffraction on cytoskeletal networks

The nano-scale structure of cytoskeletal biopolymers as well as sophisticated superstructures determine the versatile cellular shapes and specific mechanical properties. One example is keratin intermediate filaments in epithelial cells, which form thick bundles that can further organize in a cross-linked network. To study the native structure of keratin bundles in whole cells, high-resolution techniques are required, which do at the same time achieve high penetration depths. We employ scanning x-ray diffraction using a nano-focused x-ray beam to study the structure of keratin in freeze-dried eukaryotic cells. By scanning the sample through the beam we obtain x-ray dark-field images with a resolution of the order of the beam size, which clearly show the keratin network. Each individual diffraction pattern is further analyzed to yield insight into the local sample structure, which allows us to determine the local structure orientation. Due to the small beam size we access the structure in a small sample volume without performing the ensemble average over one complete cell.

Uncorrelated Local Maximum Margin Criterion: An Efficient Dimensionality Reduction Method for Text Classification

The pragmatic realism of the high dimensionality incurs limitations in many pattern recognition arena such as text classification,data mining, information retrieval and face recognition. The unsupervised PCA ante up no attention to the class labels of the existing training data. LDA is not stable due to the small sample size problem and but corroborate best directions when each class has a Gaussian density with a common covariance matrix. But it can flop if the class densities are more general and interpreted class separability in between-class-matrices are inadequate. Maximum Margin Criterion (MMC) having lower computational cost, is more efficient than LDA for calculating the discriminant vectors barring the computation for inverse of within-class-scatter matrix. But traditional MMC disregards the discriminative information within the local structure of samples and performance is depended on choosing of a coefficient. In this paper we delineate the locality of data points by counting a distances among data points considering the supervised knowledge. We have computed the entire scatter matrix in Laplacian graph embedded space and finally the produced stastically uncorrelated discriminant vectors reduces redundancy among the extracted features and there is no constant to be chosen. Our experiment with Reauter dataset recommends this algorithm is more efficient than LDA, MMC and it manifests similar or sometimes better result than other locality based algorithm like LPP and LSDA.

Mixture graph based semi-supervised dimensionality reduction

Graph structure is crucial to graph based dimensionality reduction. A mixture graph based semi-supervised dimensionality reduction (MGSSDR) method with pairwise constraints is proposed. MGSSDR first constructs multiple diverse graphs on different random subspaces of dataset, then it combines these graphs into a mixture graph and does dimensionality reduction on this mixture graph. MGSSDR can preserve the pairwise constraints and local structure of samples in the reduced subspace. Meanwhile, it is robust to noise and neighborhood size. Experimental results on facial images feature extraction demonstrate its effectiveness.

Feature extraction based on Laplacian bidirectional maximum margin criterion

Maximum margin criterion (MMC) based feature extraction is more efficient than linear discriminant analysis (LDA) for calculating the discriminant vectors since it does not need to calculate the inverse within-class scatter matrix. However, MMC ignores the discriminative information within the local structures of samples and the structural information embedding in the images. In this paper, we develop a novel criterion, namely Laplacian bidirectional maximum margin criterion (LBMMC), to address the issue. We formulate the image total Laplacian matrix, image within-class Laplacian matrix and image between-class Laplacian matrix using the sample similar weight that is widely used in machine learning. The proposed LBMMC based feature extraction computes the discriminant vectors by maximizing the difference between image between-class Laplacian matrix and image within-class Laplacian matrix in both row and column directions. Experiments on the FERET and Yale face databases show the effectiveness of the proposed LBMMC based feature extraction method.

Nanoscale Imaging of CaCu<sub>3</sub>Ti<sub>4</sub>O<sub>12</sub> Dielectric Properties: The Role of Surface Defects

Scanning probe microscopy with conductive tips has been used to image the dielectric properties of ceramics with giant permittivity. In particular, measurements in impedance mode and of local resistivity allowed to image the permittivity map on polycrystalline materials. Such imaging allows to correlate the dielectric properties with the local sample structure and with defects inside the single grains of the polycrystalline ceramics. However, artifacts due to surface imperfections should be distinguished from bulk properties and eliminated.

Spectroscopic study on iron doped silica-bismuthate glasses and glass ceramics

Fourier transform infrared (FTIR) and electron paramagnetic resonance (EPR) spectroscopic data obtained for iron doped silica-bismuthate glasses were used to investigate the changes induced in the local structure of samples as the ratio between Bi2O3 and SiO2 content changes from 3 to 0.66. The environment of constituent cations was investigated both in vitreous and partially crystallized samples of same composition. Progressive substitution of Bi2O3 by SiO2 contributes to the structural relaxation of vitreous network and enhances glass stability. By crystallization heat treatment the structural units appear to be more uniform as it results from the narrowing of corresponding IR bands and EPR lines. FTIR data show that more affected by composition and heat treatment is the environment of bismuth than that of silicon. EPR data indicate an ordering tendency with SiO2 content in glasses, while in vitroceramics the iron environment is much uniform in samples with less SiO2 content, where Bi12SiO20 phase is developed.