Google Inc.

Avinatan Hassidim

Yossi Matias

Dan Raz

Michal Segalov

William T Freeman

Michael Rubinstein

Kevin William Wilson

Ari J Leichtberg

Inbar Mosseri

Tali Dekel

Alon Ziv

Tzvi Itzhak Barenholz

Ariel Ephrat

Oran Lang

Marcel Mordechay Yung

Alok Kumar

Marcel M M Yung

Emilie Jeanne Anne Danna

Ron Merom

Sigalit Bar

Tzvika Hartman

Alon Mittelman

Hilly Meerovitch

Dotan Emanuel

Eyal Soha

Ido Feldman

Tal Uziely

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for audio-visual speech separation. A method includes: obtaining, for each frame in a stream of frames from a video in which faces of one or more speakers have been detected, a respective per-frame face embedding of the face of each speaker; processing, for each speaker, the per-frame face embeddings of the face of the speaker to generate visual features for the face of the speaker; obtaining a spectrogram of an audio soundtrack for the video; processing the spectrogram to generate an audio embedding for the audio soundtrack; combining the visual features for the one or more speakers and the audio embedding for the audio soundtrack to generate an audio-visual embedding for the video; determining a respective spectrogram mask for each of the one or more speakers; and determining a respective isolated speech spectrogram for each speaker.

Audio-visual speech separation

Methods and systems are provided for concealing identifying data that may be used to identify a beacon or device in broadcasts unless an observer device is able to directly or indirectly, via an authorized resolver device, translate an encrypted broadcast into the identifiable information. The wireless security scheme disclosed herein also pertains to resolving the concealed data messages to obtain the identifiable information.

Security for wireless broadcasts

Systems and methods are provided for allocating resources of a network among a plurality of traffic demands to optimize fairness and network utilization. Methods based on flow-increase dynamics converge toward an upward max-min fair (UMMF) allocation, in which the value of each traffic demand cannot be increased, along any of its paths, even if larger traffic demands are removed from the network. An efficient iterative algorithm that converges to a UMMF solution is also provided. The described methods and systems can be implemented efficiently, distributively, and asynchronously.

Iterative max-min fairness algorithms

Systems and methods for fault detection in large scale networks are provided. Probing instructions are generated by a probing controller associated with a network having a plurality of nodes. The probing instructions include a specified source node and a specified destination node. Each probing instruction is transmitted to a probing agent coupled to the specified source node and a data packet is transmitted from the probing agent to the specified destination node. A fault detection module is informed of probing instructions associated with failed transmissions, identifies a plurality of nodes having a likelihood of being in a network path associated with failed transmissions, and processes the plurality of nodes having a likelihood of being in the network paths associated with failed transmissions to identify of a set of likely failed nodes.

Systems and methods for fault detection in large scale networks

Systems and methods for splitting a network traffic flow in a data network are described. A flow of traffic between a source node and a destination node in the data network is split into a set of data paths. A data path includes one or more data links between nodes in the data network. A submap of the data network that excludes at least one data link is used to determine the set of flow paths.

Splitting a network traffic flow

Implementations of the present disclosure include providing a suggested social circle includes generating a list of contacts who, based on first predetermined criteria, are to be considered for potential inclusion in a social circle to be suggested to a particular user and generating an edge and vertex-weighted graph representation that indicates a strength of the particular user's affinity to each of the contacts. A set of vertices in the graph that has a high overall score based on predetermined factors is identified. After determining that the contacts corresponding to the identified set of vertices should be presented as a suggested social circle to the particular user, social circle data corresponding to the suggested social circle is transmitted, to a client device associated with the particular user, to display a representation of the suggested social circle on the client device.

Generating suggestions for user groups in social networks

Implementations of the present disclosure include generating one or more suggested social circles based on an analysis of past user interactions. The user can accept the suggested social circle and use it to distribute digital content. Generating a suggested social circle can include determining, for each of various sets of contacts, an initial score that represents a strength of the particular user's affinity to the set as a whole based on prior interactions of the particular user with the set, and determining, for each member of each set, an individual score that represents an average strength of the particular user's affinity to the member based on the interactions. An overall score for each set is calculated based on the initial score for the set and the individual scores of each member in the set. Based on the overall scores, at least one of the sets is identified for presentation as a suggested social circle to the particular user. Social circle data corresponding to the suggested social circle is transmitted to display a representation of the suggested social circle on a client device associated with the particular user.

Growing community of inventors

Petah Tikva, Israel

Avinatan Hassidim