Back to results
US 12,632,882 B2Granted
Compact data representation
AssigneeOracle International Corporation
InventorKarempudi V. Ramarao
Life of this patent
Estimating the term…
Abstract
Techniques are disclosed for transforming feature vectors to a compact form that improves the efficiency and speed of computing analytical operations. A system may generate feature vectors that correspond to a set of entities and select a subset of the feature vectors to act as reference vectors. The system executes a first stage of data condensation by representing each feature vector in terms of a distance value to each of the reference vectors thereby reducing a number of attributes in each vector to the number of selected reference vectors. The system may further condense the data by characterizing the distance values for the remaining vectors relative to each particular reference vector in terms of statistical characteristics of the distance value distribution.
Claims20
One or more non-transitory computer-readable media storing instructions, which when executed by one or more hardware processors, cause performance of operations comprising:
- generating a compact representation of a plurality of entities for training a machine learning model, wherein generating the compact representation comprises:
- generating a plurality of vectors representing the plurality of entities, wherein values of vectors of the plurality of vectors represent attributes corresponding to entities of the plurality of entities;
- selecting a subset of reference vectors, from the plurality of vectors, representing a subset of reference entities from the plurality of entities;
- generating a plurality of summary vectors that (a) comprises at least an order of magnitude fewer vectors than the plurality of vectors and (b) accurately represents the plurality of entities, at least by performing, for each reference vector of the subset of reference vectors:
- computing distance values between each non-reference vector selected from the plurality of vectors and a particular reference vector;
- and
- condensing the distance values at least by:
- generating statistical characteristics representing the distance values in relation to the particular reference vector;
- and
- based on the statistical characteristics, generating a summary vector representing the distance values in relation to the particular reference vector;
- wherein the compact representation of the plurality of entities comprises the plurality of summary vectors representing the distance values in relation to each of the subset of reference vectors;
- and
- training the machine learning model using the compact representation of the plurality of entities instead of the plurality of vectors, wherein training the machine learning model comprises:
- generating a training set comprising (a) the plurality of summary vectors representing the distance values in relation to each of the subset of reference vectors and (b) statistics associated with marketing campaigns targeting the plurality of entities;
- and
- training the machine learning model using the training set.
13 more elements
A method comprising:
- generating a compact representation of a plurality of entities for training a machine learning model, wherein generating the compact representation comprises:
- generating a plurality of vectors representing the plurality of entities, wherein values of vectors of the plurality of vectors represent attributes corresponding to entities of the plurality of entities;
- selecting a subset of reference vectors, from the plurality of vectors, representing a subset of reference entities from the plurality of entities;
- generating a plurality of summary vectors that (a) comprises at least an order of magnitude fewer vectors than the plurality of vectors and (b) accurately represents the plurality of entities, at least by performing, for each reference vector of the subset of reference vectors:
- computing distance values between each non-reference vector selected from the plurality of vectors and a particular reference vector;
- and
- condensing the distance values at least by:
- generating statistical characteristics representing the distance values in relation to the particular reference vector;
- and
- based on the statistical characteristics, generating a summary vector representing the distance values in relation to the particular reference vector;
- wherein the compact representation of the plurality of entities comprises the plurality of summary vectors representing the distance values in relation to each of the subset of reference vectors;
- and
- training the machine learning model using the compact representation of the plurality of entities instead of the plurality of vectors, wherein training the machine learning model comprises:
- generating a training set comprising (a) the plurality of summary vectors representing the distance values in relation to each of the subset of reference vectors and (b) statistics associated with marketing campaigns targeting the plurality of entities;
- and
- training the machine learning model using the training set;
- wherein the method is performed by at least one device including a hardware processor.
14 more elements
A system comprising:
- at least one device including a hardware processor;
- the system being configured to perform operations for generating a compact representation of a plurality of entities for training a machine learning model, the operations comprising:
- generating a compact representation of a plurality of entities for training a machine learning model, wherein generating the compact representation comprises:
- generating a plurality of vectors representing the plurality of entities, wherein values of vectors of the plurality of vectors represent attributes corresponding to entities of the plurality of entities;
- selecting a subset of reference vectors, from the plurality of vectors, representing a subset of reference entities from the plurality of entities;
- generating a plurality of summary vectors that (a) comprises at least an order of magnitude fewer vectors than the plurality of vectors and (b) accurately represents the plurality of entities, at least by performing, for each reference vector of the subset of reference vectors:
- computing distance values between each non-reference vector selected from the plurality of vectors and a particular reference vector;
- and
- condensing the distance values at least by:
- generating statistical characteristics representing the distance values in relation to the particular reference vector;
- and
- based on the statistical characteristics, generating a summary vector representing the distance values in relation to the particular reference vector;
- wherein the compact representation of the plurality of entities comprises the plurality of summary vectors representing the distance values in relation to each of the subset of reference vectors;
- and
- training the machine learning model using the compact representation of the plurality of entities instead of the plurality of vectors, wherein training the machine learning model comprises:
- generating a training set comprising (a) the plurality of summary vectors representing the distance values in relation to each of the subset of reference vectors and (b) statistics associated with marketing campaigns targeting the plurality of entities;
- and
- training the machine learning model using the training set.
15 more elements
Dependencies are inferred from the claim text.