Google Inc.

David Marwood

Shumeet Baluja

Michele Mae Covell

Michael Chu

Rohan Seth

Joshua J Sacks

Monika Hildegard Henzinger

Stephen Lacy

Nicholas M Johnston

Kyle Kakligian

Shahid Choudhry

Sergey Ioffe

Christopher Richard Uhlik

Aaron Lee

John Piscitello

Richard Tucker

Steve Okamoto

Peter C Chane

Matthew Vosburgh

Hong Guo

Aaron David Yonas

Daniel M Keysers

Gheorghe Postelnicu

Pascal Massimino

Rahul Sukthankar

The present disclosure provides systems and methods that train and use neural networks that can be run with no multiplications and no floating point operations. In particular, according to one aspect of the present disclosure, the respective non-linear and continuous activation functions typically used by the nodes of a neural network can be replaced with custom activation functions that output one of a discrete number of activation values. Likewise, according to another aspect of the present disclosure, the neural network can be trained such that each of its weights equals one of a discrete number of weight values. Taken together, this enables replacement of the typical multiplication process associated with computing a node of the network with a simple, and much faster, lookup process. In particular, a lookup table can store the result of multiplying each unique pair of activation value and weight value.

Deep neural networks with no multiplications and no floating point operations

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing a network input using a neural network to generate a network output for the network input. One of the methods includes maintaining, for each of the plurality of neural network layers, a respective look-up table that maps each possible combination of a quantized input index and a quantized weight index to a multiplication result; and generating a network output from a network input, comprising, for each of the neural network layers: receiving data specifying a quantized input to the neural network layer, the quantized input comprising a plurality of quantized input values; and generating a layer output for the neural network layer from the quantized input to the neural network layer using the respective look-up table for the neural network layer.

Look-up table based neural networks

An encoder system can include a pixel grid generator to receive an image having a first dimension, generate a grid having a second dimension, add a plurality of points to positions on the grid, and map a plurality of pixels of the image to the plurality of points. The encoder system can include a color module to assign a color to each of the plurality of points using a color table, a triangulation module to generate a plurality of vertices based on the plurality of points and triangulate the grid using the vertices, and a compression module to compress the vertices as a set of compressed vertex positions and a set of vertex colors.

Image compression and decompression using triangulation

Encoding and decoding occupancy information is disclosed. A method includes determining row sums for the region, determining column sums for the region, encoding, in a compressed bitstream, at least one of the row sums and the column sums, and encoding, in the compressed bitstream and based on a coding order, at least one of the rows and the columns of the region. The coding order is based on the encoded at least one of the row sums and the column sums. The row sums include, for each row of the region, a respective count of a number of locations in the row having a specified value. The column sums include, for each column of the region, a respective count of a number of locations in the column having the specified value. A location having the specified value is indicative of the occupancy information at the location.

Compression of occupancy or indicator grids

Methods, systems, and computer program products are provided for determining content items for delivery to users based on imagery. One example method includes determining a location of a user or a location associated with a query submitted by the user, identifying imagery associated with the determined location, and determining one or more content items for delivery to the user based at least in part on the imagery.

Determining content for delivery based on imagery

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, including a method that comprises: determining a target geographic feature that has insufficient targeting information associated therewith, the target geographic feature defining a location; determining one or more similar geographic features to the target geographic feature, each geographic feature including targeting information, the determining based on shared excess queries that are shared between geographic features that are determined to be similar to the target geographic feature, wherein each excess query is a query associated with and exceeds an expected query count for each of the similar geographic features and the target geographic feature; attributing targeting information associated with at least one of the one or more similar geographic features to the target geographic feature; and serving content responsive to queries that relate to the target geographic feature based at least in part on the attributed targeting information.

Determining labels from similar geographic features

In one aspect, a computer-implemented method for adjusting pixel saturation may generally include accessing, by one or more computing devices, a target distribution function associated with at least one target image and an input distribution function associated with at least one input image. The target distribution function may define a target probability for a pixel saturation of each pixel within the target image(s). The input distribution function may define an input probability for an initial saturation value of each pixel within the input image(s), with the input image(s) differing from the target image(s). The method may also include associating, by the computing device(s), the initial saturation value of each pixel within the input image(s) with a target saturation value based on the input and target distribution functions and adjusting, by the computing device(s), the initial saturation value of each pixel within the input image(s) to the corresponding target saturation value.

System and method for adjusting pixel saturation

Methods, systems, and apparatus, including computer programs encoded on a computer-readable storage medium, including a method for ranking content using location-based query log analysis. The method comprises: identifying a region defining an area of interest including identifying a plurality of content items that are associated with the region; evaluating query logs associated with users that submitted queries associated with the region to determine a ranking associated with the plurality of content items; receiving a request for content associated with the region; and providing one or more of the content items based at least in part on the ranking.

Ranking content using location-based query log analysis

An embodiment for reusing data in content files includes receiving a request for a content file and retrieving a recipe that includes a series of instructions needed to compose the content file. Data identified by the recipe may then be retrieved from a content file server and the requested content file is composed based on the set of instructions in the recipe. In an example, the recipe is processed by a recipe player to compose a content file. In another embodiment, content files are stored in a cache as a series of instructions for generating them from portions of other content files. In this way, performance in a content delivery network may be improved by reusing similar data in content files.

Reusing data in content files

Methods, systems, and apparatus, including computer programs encoded on a computer-readable storage medium, for providing a method that comprises: determining excess queries over multiple time periods for a given geographic feature, where the geographic feature defines a location; comparing geographic features for similarity based at least in part on the excess queries associated with a respective geographic feature; and for a given target geographic feature, determining one or more similar geographic features based on the comparing.

Growing community of inventors

San Francisco, CA, United States of America

David Marwood