MODEL TRAINING METHOD AND APPARATUS
    1.
    发明公开

    公开(公告)号:US20240143977A1

    公开(公告)日:2024-05-02

    申请号:US18496177

    申请日:2023-10-27

    CPC classification number: G06N3/045 G06N3/084

    Abstract: This application discloses a model training method, which may be applied to the field of artificial intelligence. The method includes: when training a first neural network model based on a training sample, determining N parameters from M parameters of the first neural network model based on a capability of affecting data processing precision by each parameter; and updating the N parameters. In this application, on a premise that it is ensured that the data processing precision of the model meets a precision requirement, because only N parameters in M parameters in an updated first neural network model are updated, an amount of data transmitted from a training device to a terminal device can be reduced.

    OPTICAL SIGNAL PROCESSING APPARATUS AND METHOD

    公开(公告)号:US20230239067A1

    公开(公告)日:2023-07-27

    申请号:US18190733

    申请日:2023-04-06

    CPC classification number: H04J14/0221 H04B10/293 H04J14/0201

    Abstract: Example optical signal processing apparatuses and methods are provided. One example apparatus includes: N light sources, a wavelength multiplexer, an optical processor, a dither application circuit, a first detection circuit, a second detection circuit, and a feedback control circuit. The light source generates a single-wavelength signal. The dither application circuit applies a dither signal to the light source. The wavelength multiplexer generates a multi-wavelength signal based on the single-wavelength signal. The first detection circuit is configured to obtain a first power signal of a signal input to the optical processor. The second detection circuit is configured to obtain a second power signal of a signal output from the optical processor. The feedback control circuit adjusts a working parameter of the optical processor based on the dither signal corresponding to the single-wavelength signal, the first power signal, and the second power signal.

    MULTI-WAVELENGTH LASER AND WAVELENGTH CONTROL METHOD

    公开(公告)号:US20220181845A1

    公开(公告)日:2022-06-09

    申请号:US17680851

    申请日:2022-02-25

    Abstract: A multi-wavelength laser and a wavelength control method are disclosed. The multi-wavelength laser includes a waveguide, a first electrode, and a second electrode. The first electrode and the second electrode are disposed on the waveguide. The first electrode is electrically isolated from the second electrode. The first electrode includes a plurality of sub-electrodes, and every two adjacent sub-electrodes are electrically isolated. The second electrode is configured to amplify an optical signal in the waveguide by loading a current. At least one sub-electrode is configured to adjust a wavelength of the optical signal in the waveguide by loading a current or a voltage.

    TEMPERATURE FEEDBACK CONTROL APPARATUS, METHOD, AND SYSTEM

    公开(公告)号:US20220006466A1

    公开(公告)日:2022-01-06

    申请号:US17448113

    申请日:2021-09-20

    Abstract: This application discloses a temperature feedback control apparatus, method. The method includes two electric switches, a feedback control unit and an optical component. A first electric switch is configured to control that only a first channel of at least two channels that correspond to the first electric switch is conducted at a moment, to feed back an optical signal of a target optical component connected to the first channel to the feedback control unit. The feedback control unit is configured to calculate temperature of the corresponding optical component based on an electrical signal converted from the optical signal, to obtain a control signal. The second electric switch is configured to control, when the first channel is conducted, that only the second channel is conducted, to transmit the control signal to the target optical component to adjust its temperature. The optical component connects to both the first and second channels.

Patent Agency Ranking