# US 2025/0157188 A1 · Measure similarities between sets of feature vectors in deep learning applications

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

- Number: US 2025/0157188 A1
- Document ID: US20250157188A1
- Status: Published application
- Kind: A1
- Application number: 18/507,890
- Filed: 2023-11-13
- Published: 2025-05-15
- Earliest claimed: 2023-11-13
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Parijat Dube; Bishwaranjan Bhattacharjee; Yulong Li; John Ronald Kender
- Web page: https://patentgrep.com/patent/US20250157188A1

## CPC

- G06V 10/761: {Proximity, similarity or dissimilarity measures}
- G06F 18/22: Matching criteria, e.g. proximity measures
- G06V 30/418: Document matching, e.g. of document images
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A method, system and apparatus for measuring similarities between sets of feature vectors, including generating a feature representation from existing sources of datasets, generating representation of metric distances of the existing sources to each other using energy distance measure from the feature representation, generating feature representation of a target dataset, generating representation of the metric distances of each of the target dataset using the energy distance measure, and providing a choice of existing sources which further extremizes a geometric content of a hypervolume described by a pseudolabel sequence.

## Claims

### Claim 1

A method for measuring similarities between sets of feature vectors, comprising:
generating a feature representation from existing sources of datasets;
generating representation of metric distances of the existing sources to each other using energy distance measure from the feature representation;
generating feature representation of a target dataset; and
generating representation of the metric distances of each of the target dataset using the energy distance measure.

### Claim 2 (depends on 1)

The method according to claim 1, further comprising providing a choice of existing sources which further extremizes a geometric content of a hypervolume described by a pseudolabel sequence.

### Claim 3 (depends on 2)

The method according to claim 2, further comprising repeating the providing of the choice until an empirically-determined stopping criterion is performed.

### Claim 4 (depends on 3)

The method according to claim 3, further comprising repeating the providing of the choice using differing extremizing criteria.

### Claim 5 (depends on 2)

The method according to claim 2, further comprising repeating the providing of the choice using differing extremizing criteria.

### Claim 6 (depends on 2)

The method according to claim 2, wherein the target data set comprises images,
wherein the energy distance measure comprises a statistical distance between two probability distributions,
wherein the generating representation of the metric distances includes generating the metric distances of the targets to each other and to the sources using the energy distance measure from the feature representation.

### Claim 7 (depends on 2)

The method according to claim 2, further comprising outputting pseudo labeling using the energy distance measure.

### Claim 8 (depends on 1)

A computer readable medium storing instructions according to the method of claim 1.

### Claim 9

A system for measuring similarities between sets of feature vectors, comprising:
a memory storing computer instructions;
a processor executing the computer instructions and configured to:
generate a feature representation from existing sources of datasets;
generate representation of metric distances of the existing sources to each other using energy distance measure from the feature representation;
generate feature representation of a target dataset; and
generate representation of the metric distances of each of the target dataset using the energy distance measure.

### Claim 10 (depends on 9)

The system according to claim 9, wherein the processor is further configured to provide a choice of existing sources which further extremizes a geometric content of a hypervolume described by a pseudolabel sequence.

### Claim 11 (depends on 10)

The system according to claim 10, wherein the processor is further configured to repeat the providing of the choice until an empirically-determined stopping criterion is performed.

### Claim 12 (depends on 11)

The system according to claim 11, wherein the processor is further configured to repeat the providing of the choice using differing extremizing criteria.

### Claim 13 (depends on 10)

The system according to claim 10, wherein the processor is further configured to repeat the providing of the choice using differing extremizing criteria.

### Claim 14 (depends on 10)

The system according to claim 10, wherein the target data set comprises images,
wherein the energy distance measure comprises a statistical distance between two probability distributions, and
wherein the generating representation of the metric distances includes generating the metric distances of the targets to each other and to the sources using the energy distance measure from the feature representation.

### Claim 15 (depends on 14)

The system according to claim 14, wherein the processor is further configured to outputting pseudo labeling using the energy distance measure.

### Claim 16

A method for measuring similarity between two sets of documents, the method comprising:
calculating one or more contextualized token vector representations for two or more documents in a database;
determining an internal energy of the two or more documents based on the one or more vector representations;
calculating one or more contextualized token vector representations of a query;
determining an internal energy of the query;
calculating a cross energy of the query and the two or more documents;
determining an energy distance based on the internal energy and cross energy of the query and the two or more documents;
ranking the query and two or more documents by similarity based on the energy distance; and
retrieving a document from the two or more documents based on the ranking.

### Claim 17 (depends on 16)

The method according to claim 16, wherein the ranking is further based on one or more distance metrics including cosine distance.

### Claim 18 (depends on 16)

The method according to claim 16, wherein the method is performed offline.

### Claim 19 (depends on 16)

A system, comprising:
a memory storing computer instructions; and
a processor executing the computer instructions comprising the method of claim 16.

### Claim 20 (depends on 16)

A computer readable medium comprising computer instruction of the method of claim 16.
