Microsoft Technology Licensing, LLC

Samsung Electronics Co., Ltd.

Zhouchen Lin

Jian Wang

Xiaoou Tang

Qiang Wang

Chunhui Zhang

Yingnong Dang

Mingqing Xu

Yue Li

Yu Zou

Jiang Wu

Xianfang Wang

Deli Zhao

Heung-Yeung Shum

Dongmei Zhang

Baining Guo

Xin Tong

Jiaping Wang

Liang Wan

Junfeng He

Chun-Hui Hu

Lei Huang

Xu Yang

Stephen S Lin

Liyong Chen

Lifeng Wang

Yantao Li

Jian-Lai Zhou

Xiaohui Hou

Guangcan Liu

Zhicheng Zhong

Rongrong Wang

Wei Zhang

Zhengyou Zhang

Xin Zhang

Jian Jun Wang

Xiao Bi Liang

Shi Han

Bin Dai

Rong Xiao

Zhengdong Zhang

Xiaoxu Ma

Yi Jie Ma

Tien-Tsin Wong

Yue Dong

Moshe Ben-Ezra

Tongxian Chen

Bennett Sampson Wilburn

Yihua Xu

Wei Feng

Haitao Chen

Xun Xu

Sijun Liu

Tianjun Ye

Zhi-Qiang Liu

Tian Fang

Xuan Yu

Yunbo Wang

Bin Wan

Liyon Chen

A neural network update method, a classification method and an electronic device thereof are provided. The neural network update method includes obtaining input data of the neural network; identifying an active neuron and an inactive neuron from among at least one neuron included in the neural network; and based on an information bottleneck of the at least one neuron, training the neural network based on the input data while maintaining information bottleneck of the active neuron.

Neural network update method, classification method and electronic device

Systems and methods for detecting annotation digital ink strokes and further associating annotation digital ink strokes with word digital ink strokes are presented. Ink strokes are captured on a writing surface and then classified as words or annotations. Annotations are then anchored to corresponding words. When words are relocated or edited on the writing surface, the anchored annotations are also relocated and may even be reshaped according to the changes in the anchored words.

Annotation detection and anchoring on ink notes

Described is a technology, such as implemented in a computational software program, by which a minimal polynomial is efficiently determined for a radical expression over the ring Z of integer numbers or the ring Q of rational numbers. The levels of the radical are grouped into a level permutation group that is used to find a level permutation set. An annihilation polynomial is found based upon the level permutation set. The annihilation polynomial is factored, and a selection mechanism selects the minimal polynomial based upon the annihilation polynomial's factors.

Computing minimal polynomials

A 'Text Rectifier' provides various techniques for processing selected regions of an image containing text or characters by treating those images as matrices of low-rank textures and using a rank minimization technique that recovers and removes image deformations (e.g., affine and projective transforms as well as general classes of nonlinear transforms) while rectifying the text or characters in the image region. Once distortions have been removed and the text or characters rectified, the resulting text is made available for a variety of uses or further processing such as optical character recognition (OCR). In various embodiments, binarization and/or inversion techniques are applied to the selected image regions during the rank minimization process to both improve text rectification and to present the resulting images of text to an OCR engine in a form that enhances the accuracy of the OCR results.

Rectification of characters and text as transform invariant low-rank textures

An exemplary method includes providing image data for an illuminated physical sample of a heterogeneous translucent material, determining one or more material properties of the material based in part on a diffusion equation where one of the material properties is a diffusion coefficient for diffusion of radiation in the material and where the determining includes a regularization term for the diffusion coefficient, mapping the one or more material properties to a virtual object volume, assigning virtual illumination conditions to the virtual object volume, and rendering the virtual object volume using the virtual illumination conditions as a boundary condition for a system of diffusion equations of the virtual object volume. Other methods, devices and systems are also disclosed.

Modeling and rendering of heterogeneous translucent materals using the diffusion equation

A 'Scene Re-Lighter' provides various techniques for using an automatically reconstructed light transport matrix derived from a sparse sampling of images to provide various combinations of complex light transport effects in images, including caustics, complex occlusions, inter-reflections, subsurface scattering, etc. More specifically, the Scene Re-Lighter reconstructs the light transport matrix from a relatively small number of acquired images using a 'Kernel Nystr&#xf6;m' based technique adapted for low rank matrices constructed from sparsely sampled images. A &#x201c;light transport kernel&#x201d; is incorporated into the Nystr&#xf6;m method to exploit nonlinear coherence in the light transport matrix. Further, an adaptive process is used to efficiently capture the sparsely sampled images from a scene. The Scene Re-Lighter is capable of achieving good reconstruction of the light transport matrix with only few hundred images to produce high quality relighting results. Further, the Scene Re-Lighter is also effective for modeling scenes with complex lighting effects and occlusions.

Light transport reconstruction from sparsely captured images

Modeling and rendering of heterogeneous translucent materials using the diffusion equation

This disclosure describes an integrated framework for class-unsupervised object segmentation. The class-unsupervised object segmentation occurs by integrating top-down constraints and bottom-up constraints on object shapes using an algorithm in an integrated manner. The algorithm describes a relationship among object parts and superpixels. This process forms object shapes with object parts and oversegments pixel images into the superpixels, with the algorithm in conjunction with the constraints. This disclosure describes computing a mask map from a hybrid graph, segmenting the image into a foreground object and a background, and displaying the foreground object from the background.

Hybrid graph model for unsupervised object segmentation

An exemplary method for extracting discriminant feature of samples includes providing data for samples in a multidimensional space; based on the data, computing local similarities for the samples; mapping the local similarities to weights; based on the mapping, formulating an inter-class scatter matrix and an intra-class scatter matrix; and based on the matrices, maximizing the ratio of inter-class scatter to intra-class scatter for the samples to provide discriminate features of the samples. Such a method may be used for classifying samples, recognizing patterns, or other tasks. Various other methods, devices, system, etc., are also disclosed.

Linear laplacian discrimination for feature extraction

Global metadata, such as a document identifier, which may be a globally unique identifier, is embedded into an embedded interactive code document by combining a first m-array and a plurality of copies of the first m-array to generate a combined m-array with encoded global metadata such that respective start positions (x, y)of the plurality of copies of the first m-array in the combined m-array are each shifted, by respective amounts that are based on respective portions of the global metadata, relative to a start position of the first m-array in the combined m-array. Global metadata may be decoded from the combined m-array by determining the respective amounts by which the plurality of copies of the first m-array are shifted, relative to the first m-array, in the combined m-array and by combining the respective amounts to produce a decoded value of the global metadata.

Global metadata embedding and decoding

Systems and methods perform Laplacian Principal Components Analysis (LPCA). In one implementation, an exemplary system receives multidimensional data and reduces dimensionality of the data by locally optimizing a scatter of each local sample of the data. The optimization includes summing weighted distances between low dimensional representations of the data and a mean. The weights of the distances can be determined by a coding length of each local data sample. The system can globally align the locally optimized weighted scatters of the local samples and provide a global projection matrix. The LPCA improves performance of such applications as face recognition and manifold learning.

Laplacian principal components analysis (LPCA)

Tensor linear Laplacian discrimination for feature extraction is disclosed. One embodiment comprises generating a contextual distance based sample weight and class weight, calculating a within-class scatter using the at least one sample weight and a between-class scatter for multiple classes of data samples in a sample set using the class weight, performing a mode-k matrix unfolding on scatters and generating at least one orthogonal projection matrix.

Tensor linear laplacian discrimination for feature extraction

Described is using semi-Riemannian geometry in supervised learning to learn a discriminant subspace for classification, e.g., labeled samples are used to learn the geometry of a semi-Riemannian submanifold. For a given sample, the K nearest classes of that sample are determined, along with the nearest samples that are in other classes, and the nearest samples in that sample's same class. The distances between these samples are computed, and used in computing a metric matrix. The metric matrix is used to compute a projection matrix that corresponds to the discriminant subspace. In online classification, as a new sample is received, it is projected into a feature space by use of the projection matrix and classified accordingly.

Classification via semi-riemannian spaces

Growing community of inventors

Beijing, China

Zhouchen Lin