公开(公告)号：US20240088913A1

公开(公告)日：2024-03-14

申请号：US18463444

申请日：2023-09-08

申请人： KOREA ADVANCED INSTITUTE OF SCIENCE AND TECHNOLOGY

发明人： Min-Soo KIM ,  Seyeon OH ,  Donghyoung HAN ,  Inju NA ,  Sungwoo PARK

IPC分类号： H03M7/30

CPC分类号： H03M7/3064 ,  H03M7/3066

摘要： Provided are a graph data compression method and apparatus. A computing device divides graph data including information about a source vertex and a destination vertex of an edge into a plurality of pieces, generates a plurality of compressed data obtained by compressing the plurality of pieces, and stores the plurality of compressed data as a file. The computing device may generate a plurality of pieces having a size equal to or less than a predefined data size by recursively dividing graph data.

公开(公告)号：US20230419553A1

公开(公告)日：2023-12-28

申请号：US18242416

申请日：2023-09-05

申请人： Microsoft Technology Licensing, LLC

发明人： Martin Jon Irwin FULLER

IPC分类号： G06T9/00 ,  G06T3/40 ,  H03M7/30

CPC分类号： G06T9/001 ,  G06T3/4007 ,  H03M7/6011 ,  H03M7/3064 ,  H03M7/3066

摘要： Data compression techniques are described for saving memory space by using fewer bits to store information while achieving high fidelity. A data set may be partitioned into a plurality of regions. Locally varying numerical ranges of data values (e.g., the minimum and maximum extents) may be determined for the plurality of regions. The data in the individual regions may be encoded using a lower number of bits as interpolation values in reference to the local extents rather than being encoded using a higher number of bits as absolute values. Where there are multiple channels of data in the regions, the number of available bits for encoding the data may be dynamically allocated per region based on the relative degrees of variance in data among the multiple channels.

公开(公告)号：US11841829B2

公开(公告)日：2023-12-12

申请号：US17383555

申请日：2021-07-23

申请人： Dell Products L.P.

发明人： Xiao Na Zhang ,  Dong Liang Huang

IPC分类号： H03M7/30 ,  G06F16/174 ,  G06N20/00

CPC分类号： G06F16/1744 ,  G06N20/00 ,  H03M7/3064 ,  H03M7/3071 ,  H03M7/6064

摘要： An information handling system includes a processor configured to process a training data file to determine an optimal data compression algorithm. The processor may also perform a compression ratio analysis that includes compressing the training data file using data compression algorithms, calculating a compression ratio associated with each of the data compression algorithms, determining an optimal compression ratio from the compression ratio associated with the each data compression algorithm; and determining a desirable data compression algorithm associated with the training data file based on the optimal compression ratio. The processor may also perform a probability analysis that includes generating a symbol transition matrix based on the desirable data compression algorithm, extracting statistical feature data based on the symbol transition matrix, and generating probability matrices based on the statistical feature data to determine the optimal data compression algorithm for each segment of a working data file.

公开(公告)号：US11791837B2

公开(公告)日：2023-10-17

申请号：US18181347

申请日：2023-03-09

申请人： TENCENT AMERICA LLC

发明人： Wei Wang ,  Wei Jiang ,  Shan Liu

IPC分类号： H03M7/00 ,  H03M7/30 ,  G06N3/02 ,  H04N19/423 ,  H04N19/119 ,  H04N19/176 ,  H04N19/146 ,  H04N19/124 ,  H04N19/11 ,  H04N19/13 ,  H04N19/70

CPC分类号： H03M7/3064 ,  G06N3/02 ,  H03M7/6005 ,  H04N19/119 ,  H04N19/124 ,  H04N19/146 ,  H04N19/176 ,  H04N19/423 ,  H04N19/11 ,  H04N19/13 ,  H04N19/70

摘要： Aspects of the disclosure provide methods and apparatuses for neural network model compression/decompression. In some examples, an apparatus for neural network model decompression includes receiving circuitry and processing circuitry. The processing circuitry decodes, from a bitstream corresponding to a representation of a neural network, at least a syntax element to be applied to multiple blocks in the neural network. Then, the processing circuitry reconstructs, from the bitstream, weight coefficients in the blocks based on the syntax element.

公开(公告)号：US11748307B2

公开(公告)日：2023-09-05

申请号：US17500246

申请日：2021-10-13

申请人： EMC IP Holding Company LLC

发明人： Uri Shabi ,  Alexei Kabishcer ,  Jonathan Volij

IPC分类号： G06F16/00 ,  G06F16/174 ,  H03M7/30 ,  G06F16/13

CPC分类号： G06F16/1744 ,  G06F16/137 ,  H03M7/3064

摘要： Technology is disclosed for selectively compressing data based on similarity of pages within the data that is to be compressed. At least one corresponding hash value is generated for each one of multiple candidate pages to be compressed. In response to the hash values generated for the candidate pages, the technology selects a set of similar candidate pages from the candidate pages. The set of similar candidate pages are a subset of the candidate pages that includes less than all the candidate pages. The set of similar candidate pages are compressed as a single unit, separately from one or more other ones of the candidate pages that were not selected to be included in the set of similar candidate pages.

公开(公告)号：US11671110B2

公开(公告)日：2023-06-06

申请号：US17081642

申请日：2020-10-27

申请人： Tencent America LLC

发明人： Wei Wang ,  Wei Jiang ,  Shan Liu

IPC分类号： H03M7/00 ,  H03M7/30 ,  G06N3/02 ,  H04N19/11 ,  H04N19/13

CPC分类号： H03M7/3064 ,  G06N3/02 ,  H03M7/6005 ,  H04N19/11 ,  H04N19/13

摘要： Aspects of the disclosure provide methods and apparatuses for neural network model compression/decompression. In some examples, an apparatus for neural network model decompression includes receiving circuitry and processing circuitry. The processing circuitry decodes, from a bitstream corresponding to a representation of a neural network, at least a syntax element to be applied to multiple blocks in the neural network. Then, the processing circuitry reconstructs, from the bitstream, weight coefficients in the blocks based on the syntax element.

公开(公告)号：US10084477B2

公开(公告)日：2018-09-25

申请号：US15551377

申请日：2016-02-17

申请人： PARSTREAM GMBH

发明人： Norbert Heusser

IPC分类号： H03M7/00 ,  H03M7/40 ,  G06F17/30 ,  H03M7/30 ,  H03M7/46

CPC分类号： H03M7/40 ,  G06F16/22 ,  H03M7/3064 ,  H03M7/4031 ,  H03M7/46

摘要： Adaptively compressing an input string (10) comprising a sequence of symbols in order to create a plurality of segment dictionaries Dm, with the steps of: generating a lookup map (110); generating a key value segment Sm,n; searching the lookup map for each symbol received in the input string (120, 130); upon detecting a symbol is not stored in the lookup map, adding the symbol by storing the symbol at a next sequential key index in the lookup map lookup map (135) and assigning a next sequential key value entry to the symbol and adding this key value to the key value segment Sm,n (150); upon detecting the symbol is stored in the lookup map, adding the corresponding key value assigned to this symbol to the next sequential entry of the key value segment Sm,n (150); wherein a new key value segment Sm,n+1 of the lookup map is generated if the number of different symbols equals the number of available key values k=2n for the opened/current key value segment Sm,n (141, 142), and where-in the lookup map is converted into a segment dictionary Dm if the maximal key value size knmax=2nmax is reached (132, 133, 134), with n being any positive integral number 1 to nmax, nmax denoting the maximal bit size, and m being any positive integral number denoting the consecutive numbering of segment dictionaries Dm.

公开(公告)号：US20180060448A1

公开(公告)日：2018-03-01

申请号：US15556626

申请日：2015-03-27

申请人： Hitachi, Ltd.

发明人： Atsushi MIYAMOTO

IPC分类号： G06F17/30 ,  G06K9/62 ,  H03M7/30

CPC分类号： G06F16/9024 ,  G06F16/00 ,  G06F16/22 ,  G06F16/2465 ,  G06F16/285 ,  G06K9/6218 ,  G06K9/6219 ,  G06K9/6892 ,  H03M7/30 ,  H03M7/3059 ,  H03M7/3064

摘要： Disclosed is a computer configured to create graph data having a vertex corresponding to a single index, an edge that links a pair of the vertices having a correlation, and an edge weight as a value of the element from the correlation matrix data having correlation values between a plurality of indices as elements, in which the correlation matrix data is acquired from the storage unit, elements of a spanning tree formed by linking vertices corresponding to indices included in the acquired correlation matrix data and an element having a value equal to or greater than a predetermined threshold value are detected, and the graph data is created on the basis of the detected elements.

公开(公告)号：US09742435B1

公开(公告)日：2017-08-22

申请号：US15188834

申请日：2016-06-21

申请人： VMware, Inc.

发明人： Arnak Poghosyan ,  Ashot Nshan Harutyunyan ,  Naira Movses Grigoryan ,  Vahe Khachikyan ,  Meruzhan Kerobyan

IPC分类号： H03M7/34 ,  H03M7/30 ,  H03M7/40 ,  H03M7/46 ,  H04L29/08

CPC分类号： H03M7/3082 ,  G06F11/34 ,  H03M7/30 ,  H03M7/3064 ,  H03M7/40 ,  H03M7/46 ,  H04L29/08072 ,  H04L29/08144 ,  H04L41/046 ,  H04L43/04 ,  H04L43/0817 ,  H04L69/04

摘要： The current document is directed to a multi-stage metric-data compression method and subsystem for compressing metric data collected and stored within distributed computing systems to facilitate computer-system management and administration. In a described implementation, metric data is partitioned into constant metric data, low-variability metric data, and high-variability metric data. High-variability metric data is compressed by identifying a set of basis metrics, or independent metrics, with respect to which a remaining set of dependent metrics can be expressed using coefficient multipliers. The high-variability metric data can then be stored as a set of independent metrics and set of coefficients, along with a small amount of additional data.

公开(公告)号：US09559721B2

公开(公告)日：2017-01-31

申请号：US15064992

申请日：2016-03-09

申请人： General Electric Company

发明人： Brian Francis Howard

IPC分类号： H03M7/30 ,  G01H1/00 ,  G06F15/02

CPC分类号： H03M7/3059 ,  G01H1/003 ,  G06F15/0275 ,  H03M7/3064

摘要： Systems and methods for compressing high-frequency signals are described in certain embodiments herein. According to certain embodiments, a high-frequency signal can be converted into a lower frequency signal so that it can be processed by one or more devices in a lower frequency infrastructure. In certain embodiments, the high-frequency signal can be compressed by certain signal conditioning components and an algorithm executed by a computer processor to at least receive a high-frequency signal, correct the high-frequency signal, determine a number of samples to be taken from the high-frequency signal (i.e., sample the high-frequency signal), store a value associated with the sampled signal, and generate a waveform that includes lower frequency content that may represent the original, high-frequency signal.

摘要翻译： 在本文的某些实施例中描述了用于压缩高频信号的系统和方法。 根据某些实施例，高频信号可以被转换成较低频率信号，使得其可以由较低频率基础设施中的一个或多个设备来处理。 在某些实施例中，高频信号可被某些信号调理部件和由计算机处理器执行的算法压缩，以至少接收高频信号，校正高频信号，确定要采集的样本数 从高频信号（即，采样高频信号），存储与采样信号相关联的值，并产生包括可表示原始高频信号的较低频率内容的波形。