Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing an input through each of a plurality of layers of a neural network to generate an output, wherein the plurality of layers comprise a convolutional layer. One of the methods includes: receiving a layer input for the convolutional layer; processing the layer input to generate a layer output for the convolutional layer, comprising determining a convolution between the layer input and a filter associated with the convolutional layer; generating a spatial weight mask for the convolutional layer by using a contextual convolution block in accordance with a set of one or more spatially sensitive mask functions defined in the contextual convolution block; and determining a weighted layer output for the convolutional layer, comprising determining a product between the spatial weight mask and the layer output of the convolutional layer.

Abstract: 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.

Abstract: Techniques for identifying groups of local features in an image and presenting advertisement information associated with stored images that match one or more features within the group of local features are described. The techniques include providing an image, identifying a region of interest in the image, providing a user-selectable link associated with the region of interest in the image, receiving a request for the region of interest in the image via the user-selectable link, and presenting advertisement information associated with a stored image that matches one or more features within the requested region of interest.

Abstract: Techniques for identifying groups of local features in an image and presenting advertisement information associated with stored images that match one or more features within the group of local features are described. The techniques include providing an image, identifying a region of interest in the image, providing a user-selectable link associated with the region of interest in the image, receiving a request for the region of interest in the image via the user-selectable link, and presenting advertisement information associated with a stored image that matches one or more features within the requested region of interest.

Abstract: A computer-implemented method for generating results for a client-requested query involves receiving a query produced by a client communication device, generating a result for the query in response to reception of the query, determining one or more predictive follow-up requests before receiving an actual follow-up request from the client device, and initiating retrieval of information associated with the one or more predictive follow-up requests, and transmitting at least part of the result to the client device, and then transmitting to the client device at least part of the information associated with the one or more predictive follow-up requests.

Abstract: A computer-implemented method for providing level-reduced tensor data having improved representation of information can include obtaining input tensor data, providing the input tensor data as input to a machine-learned discretization level reduction model configured to receive tensor data having a number of discretization levels and produce, in response to receiving the tensor data, level-reduced tensor data having a reduced number of discretization levels, and obtaining, from the machine-learned discretization level reduction model, the level-reduced tensor data. The machine-learned discretization level reduction model is trained using reconstructed input tensor data generated using an output of the machine-learned discretization level reduction model. The machine-learned discretization level reduction model can include one or more level reduction layers configured to receive input having a first number of discretization levels and to provide a layer output having a reduced a number of discretization levels.

Abstract: In accordance with some embodiments of the disclosed subject matter, mechanisms for seamless audio melding between audio items in a playlist are provided.

Abstract: 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.

Abstract: In accordance with some embodiments of the disclosed subject matter, mechanisms for seamless audio melding between audio items in a playlist are provided.

Abstract: 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.

Abstract: Methods, systems and apparatus, including computer programs encoded on a computer storage medium for selecting keywords for resources are disclosed. In one aspect, a search query is received associated with a first user. A determination is made that the first user is a follower of an entity feed that is provided by a first entity and that is provided through a social network. A content item is selected having distribution parameters specifying that the content item is to be provided to users that are followers of the entity feed and that submit the search query. The selected content item is provided for the first user.

Abstract: The present disclosure is directed to encoding images. In particular, one or more computing devices can receive data representing one or more machine learning (ML) models configured, at least in part, to encode images comprising objects of a particular type. The computing device(s) can receive data representing an image comprising one or more objects of the particular type. The computing device(s) can generate, based at least in part on the data representing the image and the data representing the ML model(s), data representing an encoded version of the image that alters at least a portion of the image comprising the object(s) such that when the encoded version of the image is decoded, the object(s) are unrecognizable as being of the particular type by one or more object-recognition ML models based at least in part upon which the ML model(s) configured to encode the images were trained.

Abstract: In accordance with some embodiments of the disclosed subject matter, mechanisms for seamless audio melding between audio items in a playlist are provided.

Abstract: Systems and techniques, including computer software, for retrieving information to a mobile device involve installing a data retrieval application on the mobile device. The data retrieval application includes instructions for presenting a structured data display on the mobile device, defining a structure of the structured data display, requesting selected hyperlinks included in the structured data display, and rendering markup language information received in response to the selected hyperlinks. A user request to retrieve data through the data retrieval application is received, and data is retrieved in response to the received user request. The retrieved data is displayed according to the structure of the structured data display, and a user can select a hyperlink in the displayed data to retrieve and render markup language information using the data retrieval application.

Abstract: In accordance with some embodiments of the disclosed subject matter, mechanisms for seamless audio melding between audio items in a playlist are provided.

Abstract: A computer-implemented method of providing personalized route information involves gathering a plurality of past location indicators over time for a wireless client device, determining a future driving objective using the plurality of previously-gathered location indicators, obtaining real-time traffic data for an area proximate to the determined driving objective, and generating a suggested route for the driving objective using the near real-time traffic data.

Abstract: Ad rendering parameters for a set of two or more ads may be determined by (a) accepting, for a set of two or more ads, ad information which includes at least one ad feature having a value that depends on ad rendering parameters, and (b) determining ad rendering parameters for at least one ad from the set of two or more ads using the accepted ad information. The act of determining ad rendering parameters may use accepted ad rendering constraints. The ad rendering constraints may include space available for rendering the ads, a footprint available for rendering the ads, and/or a maximum number of ads permitted to be rendered. The act of determining ad rendering parameters may include maximizing a value associated with serving at least one ad from the set of two or more ads with ad rendering parameters subject to the ad rendering constraints. The ad rendering parameters may include sizes of the served ads, and/or a layout of the served ads.

Abstract: The present disclosure provides systems and methods for hiding information using deep neural networks. In one example, a computer-implemented method is provided to train neural networks for hiding images, which includes inputting a package image and a cover image into an image hiding neural network and generating a carrier image as an output, the carrier image comprising the package image hidden within the cover image. The method includes inputting the carrier image into an image decoding neural network and generating a reconstruction of the package image as an output. The method includes simultaneously training the image decoding neural network based at least in part on a first loss function that describes a difference between the package image and the reconstruction of the package image and the image hiding neural network based at least in part on the first loss function and on a second loss function that describes a difference between the cover image and the carrier image.

Abstract: 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.

Abstract: In accordance with some embodiments of the disclosed subject matter, mechanisms for seamless audio melding between audio items in a playlist are provided.