# US 2021/0081817 A1 · Enhanced model updating using vector space transformations for model mapping

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0081817 A1
- Document ID: US20210081817A1
- Status: Published application
- Kind: A1
- Application number: 16/818,651
- Filed: 2020-03-13
- Published: 2021-03-18
- Earliest claimed: 2019-09-13
- Assignees: Oracle International Corporation
- Applicants: Oracle International Corporation
- Inventors: Guodong Chen; Shekhar Agrawal
- Web page: https://patentgrep.com/patent/US20210081817A1

## CPC

- G06N 3/08: Learning methods
- G06N 5/04: Inference or reasoning models
- G06N 3/045: Combinations of networks
- G06N 3/0499: Feedforward networks
- G06N 3/09: Supervised learning
- G06N 20/00: Machine learning
- G06N 5/022: Knowledge engineering; Knowledge acquisition
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

The present disclosure relates to systems and methods for updating static machine-learning models (e.g., a Doc2Vec model) without needing to retrain the models. More particularly, the present disclosure relates to systems and methods that can be used to add new data to a base model by training a client model using the new data, and transforming the vector space of the client model to align with the vector space of the base model. The base model can then be updated using the realigned client model. As such, the base model can be updated with the new data without needing to retrain the base model, which can be burdensome to processing resources, insecure, and time consuming.

## Claims

### Claim 1

A computer-implemented method comprising:
accessing a first trained model associated with a first entity, the first trained model having been trained using one or more training data sets, each training data set of the one or more training data sets representing a first set of documents, and the first trained model including N dimensions in a first vector space;
accessing a second trained model associated with a second entity, the second trained model having been trained a second set of documents, and the second trained model including the N dimensions in a second vector space;
determining a set of overlapping words between the first set of documents and the second set of documents;
generating a rotation model using each of the first trained model and the second trained model, the rotation model being configured to transform the second trained model, such that the N dimensions in the second vector space are aligned with the N dimensions in the first vector space;
transforming the second trained model using the rotation model, the transformed second trained model representing the N dimensions of the second trained model in the first vector space; and
updating the first trained model using the transformed second trained model.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1, wherein generating the rotation model further comprises:
generating a transpose of the first trained model;
determining a product of the transpose of the first trained model and the second trained model;
inputting the product into a third trained model; and
generating an output in response to inputting the product into the third trained model, the output corresponding to the rotation model.

### Claim 3 (depends on 2)

The computer-implemented method of claim 2, wherein updating the first trained model further comprises:
transforming the second trained model by multiplying the rotation model with the second trained model, wherein transforming the second trained model realigns the N dimensions of the second trained model from the second vector space to the first vector space; and
for each word of the set of overlapping words:
combining a first value of each dimension of the N dimensions of the first trained model with a second value of the corresponding dimension of the N dimensions of the second trained model; and
updating the first value with the combined first value and second value.

### Claim 4 (depends on 2)

The computer-implemented method of claim 2, wherein the first trained model is generated by inputting the first set of documents into a document-to-vector model, wherein the second trained model is generated by inputting the second set of documents into the document-to-vector model, and wherein the third trained model is a singular value decomposition (SVD) model.

### Claim 5 (depends on 1)

The computer-implemented method of claim 1, wherein each of the first trained model and the second trained model includes one or more N-dimensional word vectors.

### Claim 6 (depends on 1)

The computer-implemented method of claim 1, wherein the first trained model is updated with the second set of documents associated with the second entity without needing to retrain the first trained model.

### Claim 7 (depends on 1)

The computer-implemented method of claim 1, further comprising:
inputting a particular resume into a document-to-vector model;
generating a vector representation of the particular resume;
inputting the vector representation of the particular resume into the updated first trained model; and
generating one or more outputs of the updated first trained model, wherein each output of the one or more outputs is used to determine a recommendation of whether or not to hire a candidate associated with the particular resume.

### Claim 8

A system, comprising:
one or more processors; and
a non-transitory computer-readable storage medium containing instructions which, when executed on the one or more processors, cause the one or more processors to perform operations including:
accessing a first trained model associated with a first entity, the first trained model having been trained using one or more training data sets, each training data set of the one or more training data sets representing a first set of documents, and the first trained model including N dimensions in a first vector space;
accessing a second trained model associated with a second entity, the second trained model having been trained a second set of documents, and the second trained model including the N dimensions in a second vector space;
determining a set of overlapping words between the first set of documents and the second set of documents;
generating a rotation model using each of the first trained model and the second trained model, the rotation model being configured to transform the second trained model, such that the N dimensions in the second vector space are aligned with the N dimensions in the first vector space;
transforming the second trained model using the rotation model, the transformed second trained model representing the N dimensions of the second trained model in the first vector space; and
updating the first trained model using the transformed second trained model.

### Claim 9 (depends on 8)

The system of claim 8, wherein the operation of generating the rotation model further comprises:
generating a transpose of the first trained model;
determining a product of the transpose of the first trained model and the second trained model;
inputting the product into a third trained model; and
generating an output in response to inputting the product into the third trained model, the output corresponding to the rotation model.

### Claim 10 (depends on 9)

The system of claim 9, wherein the operation of updating the first trained model further comprises:
transforming the second trained model by multiplying the rotation model with the second trained model, wherein transforming the second trained model realigns the N dimensions of the second trained model from the second vector space to the first vector space; and
for each word of the set of overlapping words:
combining a first value of each dimension of the N dimensions of the first trained model with a second value of the corresponding dimension of the N dimensions of the second trained model; and
updating the first value with the combined first value and second value.

### Claim 11 (depends on 9)

The system of claim 9, wherein the first trained model is generated by inputting the first set of documents into a document-to-vector model, wherein the second trained model is generated by inputting the second set of documents into the document-to-vector model, and wherein the third trained model is a singular value decomposition (SVD) model.

### Claim 12 (depends on 8)

The system of claim 8, wherein each of the first trained model and the second trained model includes one or more N-dimensional word vectors.

### Claim 13 (depends on 8)

The system of claim 8, wherein the first trained model is updated with the second set of documents associated with the second entity without needing to retrain the first trained model.

### Claim 14 (depends on 8)

The system of claim 8, wherein the operations further comprise:
inputting a particular resume into a document-to-vector model;
generating a vector representation of the particular resume;
inputting the vector representation of the particular resume into the updated first trained model; and
generating one or more outputs of the updated first trained model, wherein each output of the one or more outputs is used to determine a recommendation of whether or not to hire a candidate associated with the particular resume.

### Claim 15

A computer-program product tangibly embodied in a non-transitory machine-readable storage medium, including instructions configured to cause a processor to perform operations including:
accessing a first trained model associated with a first entity, the first trained model having been trained using one or more training data sets, each training data set of the one or more training data sets representing a first set of documents, and the first trained model including N dimensions in a first vector space;
accessing a second trained model associated with a second entity, the second trained model having been trained a second set of documents, and the second trained model including the N dimensions in a second vector space;
determining a set of overlapping words between the first set of documents and the second set of documents;
generating a rotation model using each of the first trained model and the second trained model, the rotation model being configured to transform the second trained model, such that the N dimensions in the second vector space are aligned with the N dimensions in the first vector space;
transforming the second trained model using the rotation model, the transformed second trained model representing the N dimensions of the second trained model in the first vector space; and
updating the first trained model using the transformed second trained model.

### Claim 16 (depends on 15)

The non-transitory machine-readable storage medium of claim 15, wherein the operation of generating the rotation model further comprises:
generating a transpose of the first trained model;
determining a product of the transpose of the first trained model and the second trained model;
inputting the product into a third trained model; and
generating an output in response to inputting the product into the third trained model, the output corresponding to the rotation model.

### Claim 17 (depends on 16)

The non-transitory machine-readable storage medium of claim 16, wherein the operation of updating the first trained model further comprises:
transforming the second trained model by multiplying the rotation model with the second trained model, wherein transforming the second trained model realigns the N dimensions of the second trained model from the second vector space to the first vector space; and
for each word of the set of overlapping words:
combining a first value of each dimension of the N dimensions of the first trained model with a second value of the corresponding dimension of the N dimensions of the second trained model; and
updating the first value with the combined first value and second value.

### Claim 18 (depends on 16)

The non-transitory machine-readable storage medium of claim 16, wherein the first trained model is generated by inputting the first set of documents into a document-to-vector model, wherein the second trained model is generated by inputting the second set of documents into the document-to-vector model, and wherein the third trained model is a singular value decomposition (SVD) model.

### Claim 19 (depends on 15)

The non-transitory machine-readable storage medium of claim 15, wherein the first trained model is updated with the second set of documents associated with the second entity without needing to retrain the first trained model.

### Claim 20 (depends on 15)

The non-transitory machine-readable storage medium of claim 15, wherein the operations further comprise:
inputting a particular resume into a document-to-vector model;
generating a vector representation of the particular resume;
inputting the vector representation of the particular resume into the updated first trained model; and
generating one or more outputs of the updated first trained model, wherein each output of the one or more outputs is used to determine a recommendation of whether or not to hire a candidate associated with the particular resume.
