Techniques to generate execution schedules from neural network computation graphs

    公开(公告)号:US11531565B2

    公开(公告)日:2022-12-20

    申请号:US16870190

    申请日:2020-05-08

    Inventor: John Brady

    Abstract: Techniques are described for a compiler scheduling algorithm/routine that utilizes backtracking to generate an execution schedule for a neural network computation graph using a neural network compiler intermediate representation of hardware synchronization counters. The hardware synchronization counters may be referred to as physical barriers, hardware (HW) barriers, or barriers and their intermediate representations may be referred to as barrier tasks or barriers. Backtracking is utilized to prevent an available number of hardware barriers from being exceeded during performance of an execution schedule. An execution schedule may be a computation workload schedule for neural network inference applications. An execution schedule may also be a first in first out (FIFO) schedule.

    METHODS AND APPARATUS TO PERFORM REMOTE MONITORING

    公开(公告)号:US20220224863A1

    公开(公告)日:2022-07-14

    申请号:US17557851

    申请日:2021-12-21

    Abstract: Methods, apparatus, systems and articles of manufacture to perform remote monitoring are disclosed. Some example methods include adjusting an image capture rate at which an image sensor captures images based on a difference image containing differences between a first image of a first set of objects and a second image of a second set of objects. Example methods also include reducing a file size of the difference image using an edge detection technique and prioritizing one or more of a set of frames based on an amount of information contained in the frames. The frames are subdivisions of the image. In further example methods, the first image is taken at a first time and the second image is taken at a second, later time, and the method includes subtracting the first image from the second image to generate the difference image.

Patent Agency Ranking