US 12,475,193 B1Granted

System to reduce data retention using aggregated embedding data

AssigneeAmazon Technologies, Inc.

InventorsAlon Shoshan, Igor Kviatkovsky, Nadav Israel Bhonker, Lior Zamir, Gerard Guy Medioni, Manoj Aggarwal, Dilip Kumar

Life of this patent

Estimating the term…

patentgrep

Loading patent…

Abstract

An image of at least a portion of a user during enrollment to a biometric identification system is acquired and processed with a plurality of models to determine embeddings that are representative of features in that image in respective embedding spaces. Transformers are trained to accept as input an embedding from a respective model and produce as output a transformed embedding consistent with a common embedding space. These transformed embeddings are aggregated to produce transformed aggregated embedding data and stored for later comparison to identify the user. Subsequently, one or more embedding models may be used to process a query image to determine a query embedding. The query embedding may be transformed into the common embedding space for comparison.

Claims20

Dependent claims

A system comprising:

a memory, storing first computer-executable instructions;
and
a hardware processor to execute the first computer-executable instructions to:
at a first time:
determine input image data that is associated with first identification data;
determine a first plurality of embedding data based on processing the input image data with each of a first plurality of embedding models;
determine a first plurality of single transform embedding data based on processing the first plurality of embedding data with a first plurality of transformer networks, wherein the first plurality of single transform embedding data is representative of a first embedding space;
determine first aggregated embedding data based on the first plurality of single transform embedding data;
store the first aggregated embedding data;
store data indicative of an association between the first aggregated embedding data and the first identification data;
and
remove the input image data from the memory before a second time;
and
at the second time:
retrieve the first aggregated embedding data;
determine query input image data;
determine query embedding data in the first embedding space based on processing the query input image data with a first embedding model;
compare the query embedding data and the first aggregated embedding data;
and
determine, based on the comparison of the query embedding data and the first aggregated embedding data, that the first identification data is associated with the query input image data.

17 more elements

A method comprising:

at a first time:
accessing image data;
determining a first plurality of embedding data by processing the image data with each of a first plurality of embedding models;
determining a first plurality of single transform embedding data by processing the first plurality of embedding data with a first plurality of transformer networks;
determining first aggregated embedding data based on the first plurality of single transform embedding data;
storing the first aggregated embedding data;
storing data indicative of an association between the first aggregated embedding data and first identification data;
and
removing the image data from memory before a second time;
and
at the second time:
retrieving the first aggregated embedding data;
determining query image data;
determining query embedding data;
comparing the query embedding data and the first aggregated embedding data;
and
determining that the first identification data is associated with the query image data.

14 more elements

A system comprising:

a memory, storing first computer-executable instructions;
and
a hardware processor to execute the first computer-executable instructions to:
at a first time:
access image data;
determine a first plurality of embedding data by processing the image data with each of a first plurality of embedding models;
determine a first plurality of single transform embedding data by processing the first plurality of embedding data with a first plurality of transformer networks;
determine first aggregated embedding data based on the first plurality of single transform embedding data;
store the first aggregated embedding data;
and
remove the image data from the memory before a second time;
and
at the second time:
retrieve the first aggregated embedding data;
determine query image data;
determine query embedding data;
compare the query embedding data and the first aggregated embedding data;
and
determine that first identification data associated with the image data is associated with the query image data.

16 more elements

Dependencies are inferred from the claim text.

Facts

Application: 17/456,055Filed 22 Nov 2021 · continuation in part
Priority: 22 Sep 2021
Granted: 18 Nov 2025
Term ends: Not estimatedThe source records don’t support an estimate.

Classification17 codes

G06F 18/21375
Based on criteria of topology preservation, e.g. multidimensional scaling or self-organising maps: involving differential geometry, e.g. embedding…
G06F 21/32
User authentication: using biometric data, e.g. fingerprints, iris scans or voiceprints
G06F 18/214
Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation: generating training…
G06F 21/6245
To a system of files or objects, e.g. local or distributed file system or database: protecting personal data, e.g. for financial or medical purposes