Google Inc.

Daniel S Tarlow

Christopher Rohrs

Adam Sah

Beverly Yang

Ryohei Takahashi

Lawrence Wen-Kai Shih

Maxim Tabachnyk

Jacob Austin

Jacob Hanson Hegna

Hassan Abolhassani

Marc Hatcher Rasi

Daniel Dun-ning Woo Johnson

Yujia Li

Matej Balog

Bart Van Merrienboer

Subhodeep Moitra

Systems and methods of the present disclosure are directed to a method for machine-learned code segment prediction for optimizing software development. The method includes obtaining an incomplete segment of code. The method includes processing the incomplete segment of code with a machine-learned code prediction model to obtain a sampled set of segment completion predictions that include code that completes the incomplete segment of code. The method includes determining an aggregated segment completion prediction from the sampled set of segment completion predictions. The method includes replacing a portion of the aggregated segment completion prediction with an input field, wherein the portion of the aggregated segment completion prediction is associated with a degree of certainty less than a threshold degree of certainty.

Machine-learned models for generating code snippets with predicted placeholders for optimizing software development

A computer-implemented method for computing node embeddings of a sparse graph that is an input of a sparse graph neural network is described. Each node embedding corresponds to a respective node of the sparse graph and represents feature information of the respective node and a plurality of neighboring nodes of the respective node. The method includes: receiving an adjacency matrix that represents edges of the sparse graph; receiving a weight matrix representing, for each node of the sparse graph, a level of influence of respective neighboring nodes on the node; initializing, for each node of the sparse graph, a respective node embedding; transforming the adjacency matrix into a low-bandwidth adjacency matrix, and performing the following operations at least once: generating a message propagation matrix as a product of the low-bandwidth adjacency matrix, the node embeddings of the nodes, and the weight matrix, wherein the message propagation matrix represents message propagation among the nodes of the sparse graph, and updating the node embeddings of the sparse graph by processing the message propagation matrix and the node embeddings of the nodes using an encoder neural network of the sparse graph neural network.

Optimizing sparse graph neural networks for dense hardware

Methods, systems, and apparatus, including computer program products, for ranking resources. One or more signals reporting one or more actions by the plurality of users related to a respective resource are received, where a reported action is a presentation of the respective resource, an addition of the respective resource, or a deletion of the respective resource. The reported actions include one or more short deletions of the respective resource. A respective long-addition click-through rate (LACTR) is determined for the respective resource based on a number of reported presentations of the respective resource and a difference between a number of reported additions of the respective resource and a number of short deletions of the respective resource. The plurality of resources are ranked based on the respective LACTRs for the plurality of resources. The ranking of the plurality of resources are provided.

Online ranking metric

Generic online ranking system and method suitable for syndication

Methods, systems, and apparatus, including computer program products, for exchanging information in accordance with a protocol. In one aspect, signals reporting user actions related to resources and ranking data ranking the resources based on the signals are exchanged in accordance with a protocol. The protocol defines types of user actions that can be reported by the signals, a format for specifying values for one or more of a set of predefined parameters for qualifying the user actions, a format for specifying one or more subscriber-specific parameters relating to the resources or the user actions, and a format for identifying resources to which the user actions are related using a namespace defined by the subscriber.

Online ranking protocol

Methods, systems, and apparatus, including computer program products, for providing ranking services. In one aspect, first signals are received from a first subscriber, and second signals are received from a second subscriber. The signals include information characterizing individual user actions relating to individual ones of first or second resources, respectively, values for one or more predefined parameters, and values for one or more first subscriber-specific parameters that are specific to the first or second subscriber, respectively. First ranking data is generated from the first signals, but not from values for any subscriber-specific parameter that is not a subscriber-specific parameter for the first subscriber. Second ranking data is generated from the second signals, but not from values for any subscriber-specific parameter that is not a subscriber-specific parameter for the second subscriber.

Growing community of inventors

Toronto, Canada

Daniel S Tarlow