Adaptive configuration of a neural network device

    公开(公告)号:US10331997B2

    公开(公告)日:2019-06-25

    申请号:US14522192

    申请日:2014-10-23

    Abstract: A first input is processed via a first configuration of a neural network to produce a first output. The first configuration defines attributes of the neural network, such as connections between neural elements of the neural network. If the neural network requires a context switch to process a second input, a second configuration is applied to the neural network to change the attributes, and the second input is processed via the second configuration of the neural network to produce a second output.

    Logical block addresses used for executing host commands

    公开(公告)号:US10025533B2

    公开(公告)日:2018-07-17

    申请号:US15401570

    申请日:2017-01-09

    Abstract: A logical block address space of a storage compute device is reserved for use in executing commands from a host. First data is received at a first portion of the logical block address space, the first data causing a computation to be performed by the storage compute device. Second data is sent to the host via a second portion of the logical block address space, the second data describing a result of the computation.

    Address mapping table compression
    15.
    发明授权

    公开(公告)号:US09946462B1

    公开(公告)日:2018-04-17

    申请号:US15044007

    申请日:2016-02-15

    Abstract: Mapping table entries that map logical block addresses to physical block addresses can be intercepted and compressed to save space. In some cases, the mapping table entries can be compressed into compression units, which can hold multiple mapping table entries. Portions of the mapping table entries can be arranged into groups, and a group can be compressed with a unique compression method. The compression method used to compress a group may be based on data characteristics of the group. When data corresponding to the mapping table entries are read or modified, the compressed data can be decompressed and provided to a requesting controller or processor. When the mapping table entry is modified, the updated mapping entry may be arranged into groups, and the groups can be compressed and stored to the compression units.

    ADAPTIVE SPANNING CONTROL
    18.
    发明申请
    ADAPTIVE SPANNING CONTROL 审中-公开
    自适应扫描控制

    公开(公告)号:US20160124846A1

    公开(公告)日:2016-05-05

    申请号:US14527013

    申请日:2014-10-29

    Abstract: The disclosed technology provides for a solid state device that adaptively determines, responsive to receipt of a write command, whether or not to partition one or more individual logical blocks of data between multiple pages of a flash storage device. According to one implementation, the partitioning (e.g., spanning) determination is based on read frequency characteristics and the internal error correction code rate of the data.

    Abstract translation: 所公开的技术提供了一种固态设备,其响应于接收到写入命令而自适应地确定是否在闪存存储设备的多个页面之间划分一个或多个单独的逻辑数据块。 根据一个实施方式,分割(例如,跨越)确定基于数据的读取频率特性和内部纠错码率。

    ADAPTIVE CONFIGURATION OF A NEURAL NETWORK DEVICE
    19.
    发明申请
    ADAPTIVE CONFIGURATION OF A NEURAL NETWORK DEVICE 审中-公开
    神经网络设备的自适应配置

    公开(公告)号:US20150324685A1

    公开(公告)日:2015-11-12

    申请号:US14522192

    申请日:2014-10-23

    CPC classification number: G06N3/04

    Abstract: A first input is processed via a first configuration of a neural network to produce a first output. The first configuration defines attributes of the neural network, such as connections between neural elements of the neural network. If the neural network requires a context switch to process a second input, a second configuration is applied to the neural network to change the attributes, and the second input is processed via the second configuration of the neural network to produce a second output.

    Abstract translation: 经由神经网络的第一配置处理第一输入以产生第一输出。 第一个配置定义了神经网络的属性,如神经网络的神经元素之间的连接。 如果神经网络需要上下文切换来处理第二输入,则将第二配置应用于神经网络以改变属性,并且通过神经网络的第二配置来处理第二输入以产生第二输出。

    Adaptive spanning control
    20.
    发明授权

    公开(公告)号:US10229055B2

    公开(公告)日:2019-03-12

    申请号:US14527013

    申请日:2014-10-29

    Abstract: The disclosed technology provides for a solid state device that adaptively determines, responsive to receipt of a write command, whether or not to partition one or more individual logical blocks of data between multiple pages of a flash storage device. According to one implementation, the partitioning (e.g., spanning) determination is based on read frequency characteristics and the internal error correction code rate of the data.

Patent Agency Ranking