Google Inc.

Sylvain Gelly

Randolph Gregory Brown

Oskar Sandberg

Olivier Jean Andre Bousquet

Neil Matthew Tinmouth Houlsby

Alexander Kolesnikov

Lucas Klaus Beyer

Xiaohua Zhai

Alexey Dosovitskiy

Mostafa Dehghani

Jakob D Uszkoreit

Dirk Weissenborn

Georg Heigold

Matthias Johannes Lorenz Minderer

Thomas Unterthiner

Hartwig Adam

Yuan Hang Li

Vivek Natarajan

Taehee Lee

Basil Mustafa

Joan Puigcerver I Perez

Yin Ching Jessica Yung

Patricia Leigh MacWilliams

Alan Prasana Karthikesalingam

Jan Freyberg

Wen Yau Aaron Loh

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network to perform a downstream computer vision task. One of the methods includes pre-training an initial neural network that shares layers with the neural network to perform an initial computer vision task and then training the neural network on the downstream computer vision task.

Transfer learning between different computer vision tasks

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images using self-attention based neural networks. One of the methods includes obtaining one or more images comprising a plurality of pixels; determining, for each image of the one or more images, a plurality of image patches of the image, wherein each image patch comprises a different subset of the pixels of the image; processing, for each image of the one or more images, the corresponding plurality of image patches to generate an input sequence comprising a respective input element at each of a plurality of input positions, wherein a plurality of the input elements correspond to respective different image patches; and processing the input sequences using a neural network to generate a network output that characterizes the one or more images, wherein the neural network comprises one or more self-attention neural network layers.

Processing images using self-attention based neural networks

Methods and apparatus for associating an entity with at least one search query. Some implementations are directed to methods and apparatus for identifying multiple queries associated with an entity and identifying one or more of the queries as an entity search query that provides desired search results for the entity. Some implementations are directed to methods and apparatus for identifying a particular entity and, in response to identifying the particular entity, identifying an entity search query corresponding to the particular entity.

Associating an entity with a search query

Methods, systems, and apparatus for scoring images related to entities. In one aspect, a method includes identifying images associated with a person, each image being included in one or more resources; obtaining, for each resource that includes one of the images, a quality score that represents a quality of the resource; for each of the images: generating an image resource quality score from the quality scores of the resources that include the image; identifying a set of similar images from the images, each similar image having a measure of similarity to the image that meets a similarity measure threshold; generating an image score based on image resource quality scores of the resources that include the similar images relative to image resource quality scores of the resources that include each of the images; and generating an image authority score based on the image resource quality score and the image score.

Scoring images related to entities

Growing community of inventors

Olten, Switzerland

Sylvain Gelly