Abstract: Systems, devices, media, and methods are presented for assembling a database for query generation. The database is assembled by receiving a query history log, the query history log including target queries and a mapping between each of the target queries and associated subqueries for each of the target queries, selecting one or more of the associated subqueries for a first target query based on a conditional probability exceeding a threshold for the associated subqueries of the first target query compared to the matching associated subqueries of the other target queries, and including the first target query and the selected one or more associated subqueries for the first target query in the in-memory data structure store for query generation.

Abstract: Systems, devices, media, and methods are presented for assembling a database for query generation. The database is assembled by receiving a query history log, the query history log including target queries and a mapping between each of the target queries and associated subqueries for each of the target queries, selecting one or more of the associated subqueries for a first target query based on a conditional probability exceeding a threshold for the associated subqueries of the first target query compared to the matching associated subqueries of the other target queries, and including the first target query and the selected one or more associated subqueries for the first target query in the in-memory data structure store for query generation.

Abstract: Methods and systems for identifying personally identifiable information (PII) are disclosed. In some aspects, frequency maps of fields storing known PII information are generated. The frequency maps may count occurrences of unique bigrams in the PII fields. A field of interest may then be analyzed to generate a second frequency map. Correlations between the first frequency maps and the second frequency map may be generated. If one of the correlations meets certain criterion, the disclosed embodiments may determine that the field of interest does or does not include PII. Access control for the field of interest may then be based on whether the field includes PII. In some aspects, a storage location of data included in the field of interest may be based on whether the field includes PII.

Abstract: Systems, devices, media, and methods are presented for assembling a database for query generation. The database is assembled by receiving a query history log, the query history log including target queries and a mapping between each of the target queries and associated subqueries for each of the target queries, selecting one or more of the associated subqueries for a first target query based on a conditional probability exceeding a threshold for the associated subqueries of the first target query compared to the matching associated subqueries of the other target queries, and including the first target query and the selected one or more associated subqueries for the first target query in the in-memory data structure store for query generation.

Abstract: Methods, computer readable media, devices, and systems provide for distributed machine learning. In one aspect, a method of training a model is disclosed. The method includes receiving, by a client device, from one or more servers, an intermediate model, training, by the client device, the intermediate model based on private data, and transmitting, by the client device, to the one or more servers, the trained intermediate model.

Abstract: Methods and systems for identifying personally identifiable information (PII) are disclosed. In some aspects, frequency maps of fields storing known PII information are generated. The frequency maps may count occurrences of unique bigrams in the PII fields. A field of interest may then be analyzed to generate a second frequency map. Correlations between the first frequency maps and the second frequency map may be generated. If one of the correlations meets certain criterion, the disclosed embodiments may determine that the field of interest does or does not include PII. Access control for the field of interest may then be based on whether the field includes PII. In some aspects, a storage location of data included in the field of interest may be based on whether the field includes PII.

Abstract: Disclosed are methods and systems for generating a suggestion inventory that provides improved user engagement while ensuring privacy of relationships on a social network. The methods and systems include accessing an entity graph that specifies connections between a plurality of users on the social network; identifying a first candidate user of the plurality of users that is indirectly connected to a given user of the plurality of users; computing a number of unique paths on the entity graph between the first candidate user and the given user; determining that the number of unique paths exceeds a first threshold and includes a minimum number of friends of the given user that are directly connected to the given user on the entity graph; and adding the first candidate friend to a friend suggestion inventory for the given user in response to the determination.

Abstract: A personalized preview system to receive a request to access a collection of media items from a user of a user device. Responsive to receiving the request to access the collection of media items, the personalized preview system accesses user profile data associated with the user, wherein the user profile data includes an image. For example, the image may comprise a depiction of a face, wherein the face comprises a set of facial landmarks. Based on the image, the personalized preview system generates one or more media previews based on corresponding media templates and the image, and displays the one or more media previews within a presentation of the collection of media items at a client device of the user.

Abstract: Systems and methods are provided for performing operations including: receiving, via a messaging application of a user device, a request to recover access to an account of a user of the messaging application; accessing a first object corresponding to a first key; receiving, from a first friend of the user on the messaging application, a second object corresponding to a first portion of a second key; receiving, from a second friend of the user on the messaging application, a third object corresponding to a second portion of the second key; deriving the second key based on the second and third objects; and recovering access to the account of the user based on the first key and the second key.

Abstract: Systems and methods are provided for performing operations comprising: storing a set of input data; generating a noise distribution based on a two-step function, wherein a height of the two-step function is determined by a privacy parameter, a width of the two-step function is determined by minimizing a variance of the noise distribution, and wherein a mean of the two-step function is determined by a value of the set of input data to be privatized; applying the noise distribution to the set of input data to generate privatized noisy output data; and transmitting the resulting privatized noisy output data in response to a request for a portion of, or a complete set of, the input data.

Abstract: Systems and methods are provided for performing operations including: receiving, via a messaging application of a user device, a request to recover access to an account of a user of the messaging application; accessing a first object corresponding to a first key; receiving, from a first friend of the user on the messaging application, a second object corresponding to a first portion of a second key; receiving, from a second friend of the user on the messaging application, a third object corresponding to a second portion of the second key; deriving the second key based on the second and third objects; and recovering access to the account of the user based on the first key and the second key.

Abstract: Systems, devices, media, and methods are presented for assembling a database for query generation. The database is assembled by receiving a query history log, the query history log including target queries and a mapping between each of the target queries and associated subqueries for each of the target queries, selecting one or more of the associated subqueries for a first target query based on a conditional probability exceeding a threshold for the associated subqueries of the first target query compared to the matching associated subqueries of the other target queries, and including the first target query and the selected one or more associated subqueries for the first target query in the in-memory data structure store for query generation.

Abstract: Systems, devices, media, and methods are presented for content searching. A query processing layer (QPL) modifies search queries prior to sending the queries to a search engine. The QPL may assist with automatic completion, spell correction, expansion (e.g., to search for symbols other than letters or search for associated terms), and translation to provide users with a better search experience. Additionally, techniques are described for automatically building QPL mappings from contemporary user searches.

Abstract: Methods, computer readable media, devices, and systems provide for distributed machine learning. In one aspect, a method of training a model is disclosed. The method includes receiving, by a client device, from one or more servers, an intermediate model, training, by the client device, the intermediate model based on private data, and transmitting, by the client device, to the one or more servers, the trained intermediate model.

Abstract: Methods, computer readable media, devices, and systems provide for distributed machine learning. In one aspect, a method of training a model is disclosed. The method includes receiving, by a client device, from one or more servers, an intermediate model, training, by the client device, the intermediate model based on private data, and transmitting, by the client device, to the one or more servers, the trained intermediate model.

Abstract: Methods and systems for identifying personally identifiable information (PII) are disclosed. In some aspects, frequency maps of fields storing known PII information are generated. The frequency maps may count occurrences of unique bigrams in the PII fields. A field of interest may then be analyzed to generate a second frequency map. Correlations between the first frequency maps and the second frequency map may be generated. If one of the correlations meets certain criterion, the disclosed embodiments may determine that the field of interest does or does not include PII. Access control for the field of interest may then be based on whether the field includes PII. In some aspects, a storage location of data included in the field of interest may be based on whether the field includes PII.

Abstract: Methods and systems for identifying personally identifiable information (PII) are disclosed. In some aspects, frequency maps of fields storing known PII information are generated. The frequency maps may count occurrences of unique bigrams in the PII fields. A field of interest may then be analyzed to generate a second frequency map. Correlations between the first frequency maps and the second frequency map may be generated. If one of the correlations meets certain criterion, the disclosed embodiments may determine that the field of interest does or does not include PII. Access control for the field of interest may then be based on whether the field includes PII. In some aspects, a storage location of data included in the field of interest may be based on whether the field includes PII.

Abstract: Systems and methods are provided for performing operations comprising: storing a set of input data; generating a noise distribution based on a two-step function, wherein a height of the two-step function is determined by a privacy parameter, a width of the two-step function is determined by minimizing a variance of the noise distribution, and wherein a mean of the two-step function is determined by a value of the set of input data to be privatized; applying the noise distribution to the set of input data to generate privatized noisy output data; and transmitting the resulting privatized noisy output data in response to a request for a portion of, or a complete set of, the input data.

Abstract: Systems and methods are provided for generating noise in user data to generate privatized data. The systems and methods generate the privatized data by performing operations comprising: storing a set of input data; generating a noise distribution based on a two-step function, wherein a height of the two-step function is determined by a privacy parameter, a width of the two-step function is determined by minimizing a variance of the noise distribution, and wherein a mean of the two-step function is determined by a value of the set of input data to be privatized; applying the noise distribution to the set of input data to generate privatized noisy output data; and transmitting the resulting privatized noisy output data in response to a request for a portion of, or a complete set of, the input data.

Abstract: Systems and methods are provided for performing operations including: receiving, via a messaging application of a user device, a request to recover access to an account of a user of the messaging application; accessing a first object corresponding to a first key; receiving, from a first friend of the user on the messaging application, a second object corresponding to a first portion of a second key; receiving, from a second friend of the user on the messaging application, a third object corresponding to a second portion of the second key; deriving the second key based on the second and third objects; and recovering access to the account of the user based on the first key and the second key.