Back to results
US 12,001,790 B2Granted
Methods for exact and approximate inverses of neural embedding models
AssigneeCapital One Services, LLC
InventorsTarek Aziz Lahlou, Nathan Wolfe, Christopher Larson
Life of this patent
Estimating the term…
Abstract
Systems, apparatuses, and methods are described for inverting neural embeddings. One or more forward neural embeddings associated with meanings, features, and/or characteristics of data samples may be generated for one or more data samples. One or more inverse neural embeddings associated with the one or more forward neural embeddings may be determined. One or more inverse feature sets for the one or more inverse neural embeddings may be generated.
Claims20
A method comprising:
- receiving, by a computing device, a plurality of data samples;
- generating a plurality of forward neural embeddings for the plurality of data samples, wherein generating a forward neural embedding, of the plurality of forward neural embeddings, for a data sample of the plurality of data samples comprises:
- determining a forward feature set for the data sample, wherein the forward feature set comprises a plurality of elements having non-negative integer values;
- and
- based on applying a dictionary to the forward feature set, determining the forward neural embedding for the data sample, wherein the forward neural embedding comprises a vector of one or more values;
- determining one or more inverse neural embeddings associated with the plurality of forward neural embeddings;
- generating one or more inverse feature sets for the one or more inverse neural embeddings, wherein generating an inverse feature set, of the one or more inverse feature sets, for an inverse neural embedding of the one or more inverse neural embeddings comprises:
- determining a target function comprising a variable corresponding to the inverse feature set for the inverse neural embedding, wherein the target function comprises a 2-norm of a result of subtracting the inverse neural embedding by a product of the dictionary and the variable;
- and
- based on minimizing the target function under a constraint that each element of a plurality of elements of the inverse feature set comprises a non-negative integer value, determining the inverse feature set for the inverse neural embedding;
- and
- causing output of the one or more inverse feature sets for the one or more inverse neural embeddings.
9 more elements
An apparatus comprising:
- one or more processors;
- and
- memory storing instructions that, when executed by the one or more processors, cause the apparatus to:
- receive a plurality of data samples;
- generate a plurality of forward neural embeddings for the plurality of data samples, wherein the instructions, when executed by the one or more processors, cause the apparatus to generate a forward neural embedding, of the plurality of forward neural embeddings, for a data sample of the plurality of data samples by:
- determining a forward feature set for the data sample, wherein the forward feature set comprises a plurality of elements having non-negative integer values;
- and
- based on applying a dictionary to the forward feature set, determining the forward neural embedding for the data sample, wherein the forward neural embedding comprises a vector of one or more values;
- determining an embedding space associated with the plurality of forward neural embeddings;
- based on selecting one or more positions within the embedding space, determining one or more inverse neural embeddings;
- generate one or more inverse feature sets for the one or more inverse neural embeddings, wherein the instructions, when executed by the one or more processors, cause the apparatus to generate an inverse feature set, of the one or more inverse feature sets, for an inverse neural embedding of the one or more inverse neural embeddings by:
- determining a target function comprising a variable corresponding to the inverse feature set for the inverse neural embedding, wherein the target function comprises a 2-norm of a result of subtracting the inverse neural embedding by a product of the dictionary and the variable;
- and
- based on minimizing the target function under a constraint that each element of a plurality of elements of the inverse feature set comprises a non-negative integer value, determining the inverse feature set for the inverse neural embedding;
- and
- cause output of the one or more inverse feature sets for the one or more inverse neural embeddings.
13 more elements
A non-transitory machine-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform steps comprising:
- receiving a plurality of data samples;
- generating a plurality of forward neural embeddings for the plurality of data samples, wherein generating a forward neural embedding, of the plurality of forward neural embeddings, for a data sample of the plurality of data samples comprises:
- determining a forward feature set for the data sample, wherein the forward feature set comprises a plurality of elements having non-negative integer values;
- and
- based on applying a dictionary to the forward feature set, determining the forward neural embedding for the data sample, wherein the forward neural embedding comprises a vector of one or more values;
- based on proximity to one or more forward neural embeddings of the plurality of forward neural embeddings, determining one or more inverse neural embeddings;
- generating one or more inverse feature sets for the one or more inverse neural embeddings, wherein generating an inverse feature set, of the one or more inverse feature sets, for an inverse neural embedding of the one or more inverse neural embeddings comprises:
- determining a target function comprising a variable corresponding to the inverse feature set for the inverse neural embedding, wherein the target function comprises a 2-norm of a result of subtracting the inverse neural embedding by a product of the dictionary and the variable;
- and
- based on minimizing the target function under a constraint that each element of a plurality of elements of the inverse feature set comprises a non-negative integer value, determining the inverse feature set for the inverse neural embedding;
- and
- causing output of the one or more inverse feature sets for the one or more inverse neural embeddings.
9 more elements
Dependencies are inferred from the claim text.