公开(公告)号：US20220253694A1

公开(公告)日：2022-08-11

申请号：US17560118

申请日：2021-12-22

Applicant: Google LLC

Inventor： Ibrahim Alabdulmohsin ,  Hartmut Maennel ,  Daniel M. Keysers

IPC: G06N3/08 ,  G06N3/04

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network using re-initialization. One of the methods includes, at each time step in a sequence of time steps: identifying current values of the weights as of the training time step; selecting one of the layer blocks; generating new values for the weights of the plurality of neural network layers, comprising: re-initializing the values of the weights of at least the neural network layers in the layer blocks that are after the selected layer block without re-initializing the current values of the weights of the neural network layers in the layer block and the neural network layers in any layer block that is before the selected layer block; and raining the neural network starting from the new values for the weights of the plurality of neural network layers.

公开(公告)号：US20240169715A1

公开(公告)日：2024-05-23

申请号：US18518075

申请日：2023-11-22

Applicant: GOOGLE LLC

Inventor： Lucas Klaus Beyer ,  Pavel Izmailov ,  Simon Kornblith ,  Alexander Kolesnikov ,  Mathilde Caron ,  Xiaohua Zhai ,  Matthias Johannes Lorenz Minderer ,  Ibrahim Alabdulmohsin ,  Michael Tobias Tschannen ,  Filip Pavetic

IPC: G06V10/82 ,  G06V10/22

CPC classification number: G06V10/82 ,  G06V10/23

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a neural network that is configured to process an input image to generate a network output for the input image. In one aspect, a method comprises, at each of a plurality of training steps: obtaining a plurality of training images for the training step; obtaining, for each of the plurality of training images, a respective target output; and selecting, from a plurality of image patch generation schemes, an image patch generation scheme for the training step, wherein, given an input image, each of the plurality of image patch generation schemes generates a different number of patches of the input image, and wherein each patch comprises a respective subset of the pixels of the input image.