中科微点-全球专利检索

  1. Login
  2. Sign Up

Search Results

11 records (took 0.014 seconds)

    Compression technique for deep neural network weights
    1.
    发明授权
    Compression technique for deep neural network weights 有权

    公开(公告)号：US11757469B2

    公开(公告)日：2023-09-12

    申请号：US17220620

    申请日：2021-04-01

    Applicant: QUALCOMM Incorporated

    Inventor： Prajakt Kulkarni ,  Lakshmi Narayana Macha ,  Haoping Xu

    IPC: H03M7/30 ,  G06N3/04

    CPC classification number: H03M7/702 ,  G06N3/04

    Abstract: Various embodiments include methods and devices for compression and decompression of weight data sets. Some embodiments may include compressing weight data by receiving a weight data set of binary numbers representing weight values, generating a frame payload including a compressed first frame of a first subset of the weight values in the weight data set, and generating a block of compressed weight data having the frame payload. Some embodiments may include decompressing weight data by retrieving a block of compressed weight data, in which the block of compressed weight data includes a frame header associated with a frame payload, in which the frame header includes a normalization factor indicator, and in which the frame payload includes compressed weight values, and generating a first decompressed frame comprising decompressed weight values of the compressed weight values of the frame payload.

    CONFIGURABLE NONLINEAR ACTIVATION FUNCTION CIRCUITS
    2.
    发明申请
    CONFIGURABLE NONLINEAR ACTIVATION FUNCTION CIRCUITS 有权

    公开(公告)号：US20230083597A1

    公开(公告)日：2023-03-16

    申请号：US17807125

    申请日：2022-06-15

    Applicant: QUALCOMM Incorporated

    Inventor： Suren Mohan ,  Ren Li ,  Prajakt Kulkarni ,  Aaron Douglass Lamb

    IPC: G06N3/04

    Abstract: Certain aspects of the present disclosure provide a processor, comprising: a configurable nonlinear activation function circuit configured to: determine, based on a selected nonlinear activation function, a set of parameters for the nonlinear activation function; and generate output data based on application of the set of parameters for the nonlinear activation function, wherein: the configurable nonlinear activation function circuit comprises at least one nonlinear approximator comprising at least two successive linear approximators, and each linear approximator of the at least two successive linear approximators is configured to approximate a linear function using one or more function parameters of the set of parameters.

    Packet-based universal bit-field masking coding using configurable sparsity information
    3.
    发明授权
    Packet-based universal bit-field masking coding using configurable sparsity information 有权

    公开(公告)号：US12132502B1

    公开(公告)日：2024-10-29

    申请号：US18483389

    申请日：2023-10-09

    Applicant: QUALCOMM Incorporated

    Inventor： Haoping Xu ,  Prajakt Kulkarni ,  Suze Balatsos ,  Zhaohui Du ,  Shiqi Sun ,  Xiaoxuan Yu ,  Nanda Kumar Aswatha Kumar ,  Sheng Zhan

    IPC: G06F16/30 ,  H03M7/30

    CPC classification number: H03M7/3066 ,  H03M7/6011 ,  H03M7/6076

    Abstract: Systems and techniques are provided for compressing data. A process can include generating a compressed sub-packet by removing one or more sparsity bytes from a sequence of values corresponding to a sub-packet, the sequence of values including one or more sparsity bytes each equal to a configured sparsity value and one or more non-sparsity bytes each corresponding to a respective data value different from the configured sparsity value. A sub-packet header can be generated for the compressed sub-packet, and indicative of a respective location within the sequence of values of each non-sparsity byte. A packet header can be generated for a plurality of compressed sub-packets, and indicative of the configured sparsity value and respective coding information for each compressed sub-packet. A compressed data packet can be generated to include at least the packet header, the sub-packet header, and the one or more non-sparsity bytes included in the sequence of values.

    Compression Technique For Deep Neural Network Weights
    5.
    发明申请
    Compression Technique For Deep Neural Network Weights 有权

    公开(公告)号：US20220321143A1

    公开(公告)日：2022-10-06

    申请号：US17220620

    申请日：2021-04-01

    Applicant: QUALCOMM Incorporated

    Inventor： Prajakt Kulkarni ,  Lakshmi Narayana Macha ,  Haoping Xu

    IPC: H03M7/30 ,  G06N3/04

    Abstract: Various embodiments include methods and devices for compression and decompression of weight data sets. Some embodiments may include compressing weight data by receiving a weight data set of binary numbers representing weight values, generating a frame payload including a compressed first frame of a first subset of the weight values in the weight data set, and generating a block of compressed weight data having the frame payload. Some embodiments may include decompressing weight data by retrieving a block of compressed weight data, in which the block of compressed weight data includes a frame header associated with a frame payload, in which the frame header includes a normalization factor indicator, and in which the frame payload includes compressed weight values, and generating a first decompressed frame comprising decompressed weight values of the compressed weight values of the frame payload.

    Systems and techniques for processing keywords in audio data
    6.
    发明授权
    Systems and techniques for processing keywords in audio data 有权

    公开(公告)号：US11269592B2

    公开(公告)日：2022-03-08

    申请号：US16794553

    申请日：2020-02-19

    Applicant: QUALCOMM Incorporated

    Inventor： Andrew Kostic ,  Prajakt Kulkarni

    IPC: G06F3/16 ,  G10L15/18 ,  G10L15/02 ,  G10L15/22 ,  G10L15/08

    Abstract: Methods, systems, and devices for systems and techniques for processing keywords in audio data are described. In some devices configured with a virtual assistant, an audio processing component may support a command-first, keyword-second voice activation procedure. The audio processing component may receive audio data from a microphone and may compress a portion of the audio data and store the compressed audio data in a first buffer and may store a portion of the audio data that is uncompressed in a second buffer. The audio processing component may use the uncompressed audio data to detect the presence of a keyword and use the compressed audio data to identify a command associated with the keyword. Upon detection of the keyword, the audio processing component may decompress the compressed audio data and transmit the decompressed audio data and the uncompressed audio data to a main processor of the device.

    Data re-encoding for energy-efficient data transfer in a computing device
    10.
    发明授权
    Data re-encoding for energy-efficient data transfer in a computing device 有权

    公开(公告)号：US11636057B2

    公开(公告)日：2023-04-25

    申请号：US17390215

    申请日：2021-07-30

    Applicant: QUALCOMM INCORPORATED

    Inventor： Engin Ipek ,  Bohuslav Rychlik ,  George Patsilaras ,  Prajakt Kulkarni ,  Can Hankendi ,  Fahad Ali ,  Jeffrey Gemar ,  Matthew Severson

    IPC: G06F13/16 ,  H03M5/14

    Abstract: The energy consumed by data transfer in a computing device may be reduced by transferring data that has been encoded in a manner that reduces the number of one “1” data values, the number of signal level transitions, or both. A data destination component of the computing device may receive data encoded in such a manner from a data source component of the computing device over a data communication interconnect, such as an off-chip interconnect. The data may be encoded using minimum Hamming weight encoding, which reduces the number of one “1” data values. The received data may be decoded using minimum Hamming weight decoding. For other computing devices, the data may be encoded using maximum Hamming weight encoding, which increases the number of one “1” data values while reducing the number of zero “0” values, if reducing the number of zero values reduces energy consumption.

Patent Agency Ranking