Microsoft Technology Licensing, LLC

Amazon Technologies, Inc.

Orbeus, Inc.

International Business Machines Corporation

Google Inc.

University of Washington

Hao Chen

Kim Laine

Meng Wang

Yi Li

Tianqiang Liu

Kim Henry Martin Laine

Yuhou Xia

Kristin E Lauter

Peter Byerley Rindal

Wei Xia

Joseph Patrick Tighe

Felix Livni

Gizem Selcan Cetin

Zhicong Huang

Jin Li

Philip Andrew Chou

Sanjeev Mehrotra

Andrea Olgiati

Cheng Huang

Sastry Duri

Vasanth Bala

Madanlal S Musuvathi

Rahul Bhotika

Gary William Hall

Mark A Ganter

Duane William Storti

Todd Douglas Mytkowicz

Colin Michael Riley

Ranju Das

Keith W Ross

Jonathan Wu

Jonathan Hedley

Chao Zhang

Saeed Maleki

Srinath T V Setty

Hao Wu

Kyoohyung Han

Sebastian G Angel

Xiang Xu

Davide Modolo

Jay Thaler

Stephen Gould

Kaustav Kundu

George M Turkiyyah

Ian Porteous

Manuel Bronstein

Surbhi Maheshwari

Zheng Zhang

Michael Quang Thai Lam

Ivann Alvarado

Enrica Maria Filippi

Olli Ilari Saarikivi

Roshan Dathathri

Manoj Gupta

Matt Van Gorder

Hao Li

Xinyu Li

Camille Claire Le Men

Jane Huang

Venkatesh Bagaria

Daniel Zink

Olli Llari Saarikivi

Techniques for liveness detection based on gesture validation, facial expression analysis, and concurrency validation. The techniques include selecting a color light pattern challenge and sending the color light pattern challenge to a personal computing device for display on a display screen of the personal computing device. A set of target images (video) is received by a liveness detection service in a provider network from the personal computing device as a response to the challenge. The liveness detection service analyzes the set of target images for macro-facial expression and micro-facial expressions. A liveness determination is made by the liveness detection service as to whether the user of the personal computing device is a live genuine user or an impersonated user based on the analysis of the macro and micro-facial expressions detected in the set of target images.

Liveness detection based on gesture validation, facial expression analysis, and concurrency validation

Systems and methods for enabling constant plaintext space in bootstrapping in fully homomorphic encryption (FHE) are disclosed. A computer-implemented method for producing an encrypted representation of data includes accessing a set of encoded digits. The method includes applying an inverse linear transformation to the set of encoded digits to obtain a first encoded polynomial. The method includes applying a modulus switching and dot product with bootstrapping key to add an error term to each of the encoded digits in the first polynomial to obtain a second encoded polynomial. The method includes applying a linear transformation to the second encoded polynomial to obtain a first batch encryption. The method includes applying digit extraction to the first batch encryption to obtain a second batch encryption, the second batch encryption corresponding to the set of encoded digits without the error term.

Enabling constant plaintext space in bootstrapping in fully homomorphic encryption

A method is disclosed for providing, for display to a content provider, a user interface comprising an option to view new audiences to be added to a plurality of users currently designated to receive content of the content provider, receiving a user selection of the option, and causing display of information identifying the new audiences, the information identifying the new audiences comprises, for each new audience, an audience identifier, an indication of an estimated number of user actions related to the content of the content provider, and an option to request that a corresponding audience be added to the plurality of users currently designated to receive the content of the content provider.

Method for identifying new audiences for content of a content provider

Techniques for user customized private label prediction are described. According to some embodiments, customers can train a classifier to detect new objects in image data. These new objects may not be included in a base model provided by a service provider system. The base model can be utilized to perform object detection and feature extraction from training images that are annotated by the customer to identify the new objects. Once trained, the new custom model can be used to identify the new objects in input images and label the images accordingly.

User customized private label prediction

Systems and methods for computing a private set intersection are disclosed. A method includes storing, at a sender device, a first set of values. The method includes receiving, from a receiver device, a homomorphic encryption of a receiver device value. The method includes computing a homomorphically encrypted number based on a difference between the homomorphic encryption of the receiver device value and each value in the first set of values, and based on a hash function of the encryption of the receiver device value. The method includes transmitting the homomorphically encrypted number to the receiver device for determination, at the receiver device, whether the receiver device value is in the first set of values.

Computing a private set intersection

Methods, systems, and computer-readable media for content moderation using object detection and image classification are disclosed. A content moderation system performs object detection on an input image using one or more object detectors. The object detection finds one or more elements in the input image. The content moderation system performs classification based at least in part on the input image using one or more image classifiers. The classification determines one or more values indicative of one or more content types in the input image. The content moderation system determines one or more scores for one or more content labels corresponding to the one or more content types. At least one of the scores represents a finding of one or more of the content types in the input image. The content moderation system generates output indicating the finding of the one or more of the content types.

Content moderation using object detection and image classification

In aspects of multiplication operations on homomorphic encrypted data, a computing device stores homomorphic encrypted data as a dataset, and implements an encryption application that can perform multiplication operations on ciphertexts in the homomorphic encrypted data, where the ciphertexts include polynomial variables of the ciphertexts. The encryption application can compute and store intermediate polynomial variables that are computed as the multiplication operations are performed. The encryption application can then utilize one or more of the intermediate polynomial variables rather than recomputing the intermediate polynomial variables as the multiplication operations are performed on the ciphertexts.

Multiplication operations on homomorphic encrypted data

Embodiments of the present invention provide systems, methods, and computer storage media directed to optimizing the generation, evaluation, and selection of tensor circuit specifications for a tensor circuit to perform homomorphic encryption operations on encrypted data. A computing device having an improved compiler and runtime configuration can obtain a tensor circuit and associated schema. The computing device can map the obtained tensor circuit to an equivalent tensor circuit, adapted to perform fully homomorphic encryption (FHE) operations, and instantiated based on the obtained associated scheme. The computing device can then monitor a flow of data through the equivalent FHE-adapted tensor circuit utilizing various tensor circuit specifications determined therefor. A cost of each tensor circuit specification can be determined by the computing device based on the monitored flow of data, so as to identify an optimal set of optimal tensor circuit specifications that can be employed by the obtained tensor circuit, to efficiently perform homomorphic encryption operations on encrypted data.

Homomorphic evaluation of tensor programs

People can be tracked across multiple segments of video data, which can correspond to different scenes in a single video file, or multiple video streams or feeds. An instance of video data can be broken up into segments that can each be analyzed to determine faces and bodies represented therein. The bodies can be analyzed across frames of the segment to determine body tracklets that are consistent across the segment. Associations of faces and bodies can be determined based using relative distances and/or spatial relationships. A subsequent clustering of these associations is performed to attempt to determine consistent associations that correspond to unique individuals. Unique identifiers are determined for each person represented in one or more segments of an instance of video data. Such an approach enables individual representations to be correlated across multiple instances.

Person tracking across video instances

The present disclosure provides for customizable content moderation using neural networks with fine-grained and dynamic image classification ontology. A content moderation system of the present disclosure may provide a plurality of image categories in which a subset of of image categories may be designated as restricted categories. The restricted categories may be chosen by a content provider or an end-user. The content moderation system may utilize a neural network to classify image data (e.g., still images, video, etc.) into one or more of the plurality of image categories, and determine that an image is a restricted image upon classifying the image into one of the restricted categories. The restricted image may by flagged, rejected, removed, or otherwise filtered upon being classified as a restricted image.

Fine-grain content moderation to restrict images

Methods according to the present disclosure include: creating a representative profile from at least one change profile with an identifying tag from a repository, the representative profile including fewer than all possible changes to a system from an event; creating a candidate profile for a first candidate system, not flagged as similar or dissimilar, from the plurality of systems, wherein the candidate profile includes fewer than all changes to the first candidate system from the event; calculating at least one difference between the representative profile and the candidate profile; where the at least one difference is not within at least one threshold, flagging the first candidate system as dissimilar; and where the at least one difference is within the at least one threshold, flagging the first candidate system as similar, associating the identifying tag with the candidate profile, and adding the candidate profile to the repository.

Creating knowledge base of similar systems from plurality of systems

Systems, methods, and computer-executable instructions for reducing amortized computational costs for a query that includes receiving at least two indexes for elements stored in an n-element database. The n-element database is encoded into at least three buckets. Each element is stored within at least two buckets. Each bucket stores a proper subset of the n-elements. For each of the two indexes, a bucket is determined to retrieve the element at the index. The determined buckets are queried to retrieve the elements. The elements at the indexes are retrieved based on the querying the determined buckets.

Private information retrieval with probabilistic batch codes

In aspects of string matching in encrypted data, a computing device stores homomorphic encrypted data as a dataset, and implements a string matching application that receives an encrypted query string as a query of the homomorphic encrypted data. The string matching application can then apply algorithms to perform addition and multiplication operations, and determine whether there are matching strings of the encrypted query string in the dataset. The string matching application can compute, for each row of the dataset, a sum of some function of dataset bits and query bits for a row result, and multiply the row results of the computed rows to determine matching strings. Alternatively, the string matching application can compute, for each row of the dataset, a product over some function of the dataset bits and the query bits for a row result, and add the row results of the computed rows to determine matching strings.

String matching in encrypted data

Growing community of inventors

Seattle, WA, United States of America

Hao Chen