Patents by Inventor Gregory David Roberts

Gregory David Roberts has filed for patents to protect the following inventions. This listing includes patent applications that are pending as well as patents that have already been granted by the United States Patent and Trademark Office (USPTO).

  • Efficient data layouts for convolutional neural networks
    Patent number: 11720800
    Abstract: Systems and methods for efficient implementation of a convolutional layer of a convolutional neural network are disclosed. In one aspect, weight values of kernels in a kernel stack of a convolutional layer can be reordered into a tile layout with tiles of runnels. Pixel values of input activation maps of the convolutional layer can be reordered into an interleaved layout comprising a plurality of clusters of input activation map pixels. The output activation maps can be determined using the clusters of the input activation map pixels and kernels tile by tile.
    Type: Grant
    Filed: November 19, 2021
    Date of Patent: August 8, 2023
    Assignee: Magic Leap, Inc.
    Inventors: Ashkan Aliabadi, Gregory David Roberts
  • EFFICIENT DATA LAYOUTS FOR CONVOLUTIONAL NEURAL NETWORKS
    Publication number: 20220076056
    Abstract: Systems and methods for efficient implementation of a convolutional layer of a convolutional neural network are disclosed. In one aspect, weight values of kernels in a kernel stack of a convolutional layer can be reordered into a tile layout with tiles of runnels. Pixel values of input activation maps of the convolutional layer can be reordered into an interleaved layout comprising a plurality of clusters of input activation map pixels. The output activation maps can be determined using the clusters of the input activation map pixels and kernels tile by tile.
    Type: Application
    Filed: November 19, 2021
    Publication date: March 10, 2022
    Inventors: Ashkan Aliabadi, Gregory David Roberts
  • Efficient data layouts for convolutional neural networks
    Patent number: 11182645
    Abstract: Systems and methods for efficient implementation of a convolutional layer of a convolutional neural network are disclosed. In one aspect, weight values of kernels in a kernel stack of a convolutional layer can be reordered into a tile layout with tiles of runnels. Pixel values of input activation maps of the convolutional layer can be reordered into an interleaved layout comprising a plurality of clusters of input activation map pixels. The output activation maps can be determined using the clusters of the input activation map pixels and kernels tile by tile.
    Type: Grant
    Filed: November 11, 2019
    Date of Patent: November 23, 2021
    Assignee: Magic Leap, Inc.
    Inventors: Ashkan Aliabadi, Gregory David Roberts
  • EFFICIENT DATA LAYOUTS FOR CONVOLUTIONAL NEURAL NETWORKS
    Publication number: 20200082215
    Abstract: Systems and methods for efficient implementation of a convolutional layer of a convolutional neural network are disclosed. In one aspect, weight values of kernels in a kernel stack of a convolutional layer can be reordered into a tile layout with tiles of runnels. Pixel values of input activation maps of the convolutional layer can be reordered into an interleaved layout comprising a plurality of clusters of input activation map pixels. The output activation maps can be determined using the clusters of the input activation map pixels and kernels tile by tile.
    Type: Application
    Filed: November 11, 2019
    Publication date: March 12, 2020
    Inventors: Ashkan Aliabadi, Gregory David Roberts
  • Efficient data layouts for convolutional neural networks
    Patent number: 10489680
    Abstract: Systems and methods for efficient implementation of a convolutional layer of a convolutional neural network are disclosed. In one aspect, weight values of kernels in a kernel stack of a convolutional layer can be reordered into a tile layout with tiles of runnels. Pixel values of input activation maps of the convolutional layer can be reordered into an interleaved layout comprising a plurality of clusters of input activation map pixels. The output activation maps can be determined using the clusters of the input activation map pixels and kernels tile by tile.
    Type: Grant
    Filed: October 3, 2017
    Date of Patent: November 26, 2019
    Assignee: Magic Leap, Inc.
    Inventors: Ashkan Aliabadi, Gregory David Roberts
  • EFFICIENT DATA LAYOUTS FOR CONVOLUTIONAL NEURAL NETWORKS
    Publication number: 20180096226
    Abstract: Systems and methods for efficient implementation of a convolutional layer of a convolutional neural network are disclosed. In one aspect, weight values of kernels in a kernel stack of a convolutional layer can be reordered into a tile layout with tiles of runnels. Pixel values of input activation maps of the convolutional layer can be reordered into an interleaved layout comprising a plurality of clusters of input activation map pixels. The output activation maps can be determined using the clusters of the input activation map pixels and kernels tile by tile.
    Type: Application
    Filed: October 3, 2017
    Publication date: April 5, 2018
    Inventors: Ashkan Aliabadi, Gregory David Roberts