Linkedin Corporation

Google Inc.

Color Labs, Inc.

Microsoft Technology Licensing, LLC

Noah Fiedel

Ganesh Ramanarayanan

Nicholas James Woods

Vincent Mallet

Jessica Cheng

Eric William Gillum

Jeremiah Harmsen

Sherry Qianhong Moore

Boris Mazniker

Christopher Olston

Junbin Teng

Chih-Wei Chen

Lan Liu

Jeremiah Joseph Harmsen

Bénigne Deprey

Qing Wu

Lisha Huang

Jeremy Nelson

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for augmenting a neural network with additional operations. One of the methods includes maintaining, by a computational graph system that manages execution of computational graphs representing neural network operations for users of the computational graph system, data specifying a plurality of pre-trained neural networks, wherein each of the pre-trained neural networks is a neural network that has been trained on training data to determine trained values of the respective parameters of the neural network; obtaining data specifying a user computational graph representing neural network operations, the user computational graph comprising a plurality of nodes connected by edges; identifying (i) an insertion point after a first node in the user computational graph and (ii) a particular pre-trained neural network from the plurality of pre-trained neural networks; and inserting a remote call node into the user computational graph.

Augmenting neural networks

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for batching inputs to machine learning models. One of the methods includes receiving a stream of requests, each request identifying a respective input for processing by a first machine learning model; adding the respective input from each request to a first queue of inputs for processing by the first machine learning model; determining, at a first time, that a count of inputs in the first queue as of the first time equals or exceeds a maximum batch size and, in response: generating a first batched input from the inputs in the queue as of the first time so that a count of inputs in the first batched input equals the maximum batch size, and providing the first batched input for processing by the first machine learning model.

Batching inputs to a machine learning model

Systems and methods for the measuring of affinity levels via passive and active interactions are provided. A first group of users are selected from a plurality of possible groups of users. Affinity levels are measured among the users of the first group, the affinity levels based at least partially on both passive and active interactions between an electronic device of a first user and electronic devices of users in the first group. The first user is added to the first group based on the affinity levels.

Measuring affinity levels via passive and active interactions

Systems and methods for the forming of user device groups are presented. In one example, logical relationship information describing logical relationships among a plurality of user devices is accessed. Scores for each of a plurality of possible groups are generated based at least partially on the logical relationship information and information about a first user device, but the scores not being based on any personally identifiable information about the first user of the first user device. A first group is selected from the plurality of possible groups based on the scores. Then the first user device is added to the first group.

Determining logical groups without using personal information

Systems and methods for the forming of user device groups are presented. In one example, logical relationship information describing logical relationships among a plurality of users is accessed. Potential membership of a first user in a first group of users is determined based on scores generated for each of a plurality of possible groups of users, the scores based on the logical relationship information. Then the first user is added to the first group of users. Content from an electronic device of the first user is received for broadcast to the first group of users. In response to the receiving of content, the scores for each of the plurality of possible groups of users are regenerated based on the content. Then the first user is added to or removed from a second group of users selected from the plurality of possible groups of users based on the regenerated scores.

Reranking of groups when content is uploaded

A system and machine-implemented method including receiving a request for a set of content items within a session, identifying, in response to the request, one or more content items, determining if the session is associated with a continuation token, wherein the continuation token provides information regarding the content items provided for display during the session, generating a continuation token for the session when the session is not associated with a continuation token, providing at least one of the one or more content items for display and updating the continuation token with information regarding the at least one of the one or more content items provided for display and a time stamp.

Sorting and pagination of content item streams

Systems and methods for the forming of user device groups are presented. In one example, potential membership of a first user in a first group of users is determined, the first group of users containing a plurality of users, by selecting the first group from a plurality of possible groups, the selecting being based on scores generated, for each of the plurality of possible groups, the scores based at least partially on the a loneliness score for the first user, the loneliness score being an estimate of how lonely the first user is and being calculated by examining past communications between an electronic device of the first user and other electronic devices via a social network service. The first user is then added to the first group of users based on the loneliness score. Content from a second user in the first group of users is received, and then the content is distributed to users in the first group of users.

Determining membership in a group based on loneliness score

A score is generated for each of a plurality of potential user groups in which to add the first user, each potential user group containing a plurality of users, the score based on a location of the first user. The first user is added to a first user group based on the score generated for the first user group. A pinning action by the first user pinning the first user to the first user group is then received. A change in location of the first mobile device is detected, and the score of each of the potential user groups is regenerated based on the changed location, causing a determination that the first user should be removed from the first user group. The first user is maintained as a member of the first user group despite the determination that the first user should be removed based on the pinning.

Pinning users to user groups

In an example embodiment, a message comprising location information indicating a geographic location of a first user device is received. Then environmental information indicating at least one physical environmental condition in the vicinity of the first user device, as detected by an environmental sensor is received. Values representing logical connection strengths between first user of the first user device and other users, the logical connection strengths based, at least partially, on the location information and at least partially on the environmental information, are calculated. Then a second user and a third user to form a group with the first user are determined, based on the values representing the logical connection strengths. A user group is formed containing the first user, the second user, and the third user. Content is received from the first user device and transmitted to users of the user group.

User device group formation

Techniques for acceleration of remote procedure calls are disclosed. Such techniques include steps of receiving a content request, the content request including at least one data request for information from a database; analyzing the received content request; and determining whether the analyzed content request includes at least one data request that can be separately executed in parallel with execution of the received content request. In response to a determination that at least one data request that can be separately executed in parallel, initiating the execution of the parallel data request; receiving the requested data in response to that data request; locally storing the received requested data; and providing the stored received requested data in response to execution of the received content request.

RPC acceleration based on previously memorized flows

Systems and methods for the forming of user device groups are presented. First, logical relationship information describing logical relationships among a plurality of users is accessed. Then potential membership of a first user of a first user device in a first group of users is determined based on the logical relationship information, by selecting the first group from a plurality of possible groups, the selecting being based on scores generated, for each of the plurality of possible groups, by a candidate scorer module, the scores based at least partially on information about the first user device, but the scores not being based on any personal information about the first user the first group of users containing a plurality of users, including one or more users who are not logically connected to the first user. Then the first user is added to the first group of users.

In an example embodiment, logical relationship information describing logical relationships between a first user and other users is recorded, the logical relationship information associated with a social networking service. Then affinity levels for each of the logical relationships between the first user and the other users are calculated. One or more of the logical relationships between the first user and one or more of the other users are displayed in a user interface, wherein the affinity level for each of the one or more logical relationships is represented graphically in the user interface.

User interface for displaying user affinity graphically

Systems and methods for the transmission of content are provided. A user group is defined and a group data structure corresponding to the user group is stored. An item of content is received and stored as a child data structure of a parent data structure of the group data structure. In response to this storage, the item of content is automatically transmitted to all other users of the user group.

Adding user to logical group based on content

Systems and methods for the forming of user device groups are presented. Logical relationship information describing logical relationships among a plurality of users is accessed and potential membership of a first user in a first group of users is determined based on the logical relationship information, by selecting the first group from a plurality of possible groups, the selecting being based on scores generated, for each of the plurality of possible groups, by a candidate scorer module, the first group of users containing a plurality of users, including one or more users who are not logically connected to the first user in the logical relationship information. Then, in response to the receiving of content, scores for each of the plurality of possible groups are generated based on the content, and the first user is added or removed from a second group of users selected from the plurality of possible groups based on the regenerated scores.

Growing community of inventors

Palo Alto, CA, United States of America

Noah Fiedel