中科微点-全球专利检索

  1. Login
  2. Sign Up

Search Results

38 records (took 0.024 seconds)

    Method and apparatus for data-free post-training network quantization and generating synthetic data based on a pre-trained machine learning model
    1.
    发明授权
    Method and apparatus for data-free post-training network quantization and generating synthetic data based on a pre-trained machine learning model 有权

    公开(公告)号：US12154030B2

    公开(公告)日：2024-11-26

    申请号：US17096734

    申请日：2020-11-12

    Applicant: Samsung Electronics Co., Ltd.

    Inventor： Yoo Jin Choi ,  Mostafa El-Khamy ,  Jungwon Lee

    IPC: G06N3/08 ,  G06F17/18 ,  G06F18/2113 ,  G06F18/214 ,  G06F18/22 ,  G06N3/045 ,  G06N7/01

    Abstract: A method for training a generator, by a generator training system including a processor and memory, includes: extracting training statistical characteristics from a batch normalization layer of a pre-trained model, the training statistical characteristics including a training mean μ and a training variance σ2; initializing a generator configured with generator parameters; generating a batch of synthetic data using the generator; supplying the batch of synthetic data to the pre-trained model; measuring statistical characteristics of activations at the batch normalization layer and at the output of the pre-trained model in response to the batch of synthetic data, the statistical characteristics including a measured mean {circumflex over (μ)}ψ and a measured variance {circumflex over (σ)}ψ2; computing a training loss in accordance with a loss function Lψ based on μ, σ2, {circumflex over (μ)}ψ, and {circumflex over (σ)}ψ2; and iteratively updating the generator parameters in accordance with the training loss until a training completion condition is met to compute the generator.

    METHOD AND APPARATUS FOR INCREMENTAL LEARNING
    4.
    发明申请
    METHOD AND APPARATUS FOR INCREMENTAL LEARNING 有权

    公开(公告)号：US20220138633A1

    公开(公告)日：2022-05-05

    申请号：US17317421

    申请日：2021-05-11

    Applicant: Samsung Electronics Co., Ltd.

    Inventor： Yoo Jin Choi ,  Mostafa El-Khamy ,  Jungwon Lee

    IPC: G06N20/20 ,  G06K9/62

    Abstract: An electronic device and method for performing class-incremental learning are provided. The method includes designating a pre-trained first model for at least one past data class as a first teacher; training a second model; designating the trained second model as a second teacher; performing dual-teacher information distillation by maximizing mutual information at intermediate layers of the first teacher and second teacher; and transferring the information to a combined student model.

    Method and apparatus for learning low-precision neural network that combines weight quantization and activation quantization
    5.
    发明授权
    Method and apparatus for learning low-precision neural network that combines weight quantization and activation quantization 有权

    公开(公告)号：US11270187B2

    公开(公告)日：2022-03-08

    申请号：US15914229

    申请日：2018-03-07

    Applicant: Samsung Electronics Co., Ltd.

    Inventor： Yoo Jin Choi ,  Mostafa El-Khamy ,  Jungwon Lee

    IPC: G06N3/04 ,  G06N3/08

    Abstract: A method is provided. The method includes selecting a neural network model, wherein the neural network model includes a plurality of layers, and wherein each of the plurality of layers includes weights and activations; modifying the neural network model by inserting a plurality of quantization layers within the neural network model; associating a cost function with the modified neural network model, wherein the cost function includes a first coefficient corresponding to a first regularization term, and wherein an initial value of the first coefficient is pre-defined; and training the modified neural network model to generate quantized weights for a layer by increasing the first coefficient until all weights are quantized and the first coefficient satisfies a pre-defined threshold, further including optimizing a weight scaling factor for the quantized weights and an activation scaling factor for quantized activations, and wherein the quantized weights are quantized using the optimized weight scaling factor.

    COMPUTING SYSTEM WITH CHANNEL ESTIMATION MECHANISM AND METHOD OF OPERATION THEREOF
    7.
    发明申请
    COMPUTING SYSTEM WITH CHANNEL ESTIMATION MECHANISM AND METHOD OF OPERATION THEREOF 有权
    具有信道估计机制的计算系统及其操作方法

    公开(公告)号：US20160080174A1

    公开(公告)日：2016-03-17

    申请号：US14553917

    申请日：2014-11-25

    Applicant: Samsung Electronics Co., Ltd.

    Inventor： Yoo Jin Choi ,  Jung Hyun Bae ,  Jungwon Lee ,  Inyup Kang

    IPC: H04L25/02 ,  H04L27/00

    CPC classification number: H04L25/022 ,  H04L5/0051 ,  H04L25/021 ,  H04L27/2613 ,  H04L27/2647

    Abstract: A computing system includes: an inter-device interface configured to receive receiver signal for communicating serving content through a communication channel; a communication unit, coupled to the inter-device interface, configured to: calculate a weighting set corresponding to a modular estimation mechanism, and generate a channel estimate based on the weighting set for characterizing the communication channel for recovering the serving content.

    Abstract translation: 计算系统包括：设备间接口，被配置为接收通过通信信道传送服务内容的接收机信号; 耦合到所述设备间接口的通信单元，被配置为：计算对应于模块化估计机制的加权集合，并且基于所述加权集合生成用于表征所述通信信道以恢复所述服务内容的信道估计。

    Method and apparatus for neural network quantization
    9.
    发明授权
    Method and apparatus for neural network quantization 有权

    公开(公告)号：US12190231B2

    公开(公告)日：2025-01-07

    申请号：US15697035

    申请日：2017-09-06

    Applicant: Samsung Electronics Co., Ltd.

    Inventor： Yoo Jin Choi ,  Mostafa El-Khamy ,  Jungwon Lee

    IPC: G06N3/08 ,  G06F7/58 ,  G06N3/047 ,  G06N3/063

    Abstract: Apparatuses and methods of manufacturing same, systems, and methods for performing network parameter quantization in deep neural networks are described. In one aspect, multi-dimensional vectors representing network parameters are constructed from a trained neural network model. The multi-dimensional vectors are quantized to obtain shared quantized vectors as cluster centers, which are fine-tuned. The fine-tuned and shared quantized vectors/cluster centers are then encoded. Decoding reverses the process.

Patent Agency Ranking