# US 2025/0156752 A1 · Data augmentation using different in-domain data

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

- Number: US 2025/0156752 A1
- Document ID: US20250156752A1
- Status: Published application
- Kind: A1
- Application number: 18/509,395
- Filed: 2023-11-15
- Published: 2025-05-15
- Earliest claimed: 2023-11-15
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Youngja Park; Weiqiu You
- Web page: https://patentgrep.com/patent/US20250156752A1

## CPC

- G06N 20/00: Machine learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A method, computer system, and a computer program product are provided for data augmentation for training an artificial intelligence (AI) engine. The technique comprises encoding an in-distribution dataset having a plurality of components and an out-of-distribution dataset also having a plurality of components. The encoding is performed using a foundation model. The techniques also comprises pairing one in-distribution component from the dataset with an out-of-distribution component from the dataset in a same class to provide a first set of paired component and pairing another in-distribution component with another out-of-distribution component in a different class using the contrastive learning model to provide a second set of paired components. The first and second set of pared components are then augmented to generate an augmented training dataset. The foundation models is adjusted by using he augmented training dataset to train the AI engine.

## Claims

### Claim 1

A method for data augmentation for training an artificial intelligence (AI) engine, comprising:
encoding an in-distribution dataset having a plurality of in-distribution components, and an out-of-distribution dataset having a plurality of in-distribution components, wherein said encoding is done using a foundation model;
pairing an in-distribution component with an out-of-distribution component in a same class and using a contrastive model to provide a first set of paired components;
pairing another in-distribution component with another out-of-distribution component in a different class using said contrastive learning model to provide a second set of paired components;
augmenting said first and second set of paired components to generate an augmented training dataset; and
adjusting said foundation model by using said augmented training dataset to train said AI engine.

### Claim 2 (depends on 1)

The method of claim 1, wherein said adjusting comprises tuning and reiteratively fine-tuning said foundation model.

### Claim 3 (depends on 1)

The method of claim 1, wherein said in-distribution dataset is a labeled training dataset designated for a target task.

### Claim 4 (depends on 3)

The method of claim 3, wherein said out-of-distribution dataset is a quasi-labeled dataset designated for a similar target task.

### Claim 5 (depends on 1)

The method of claim 1, wherein said in-distribution component and said out-of-distribution component are an in-distribution sentence and an out-of-distribution sentence respectively.

### Claim 6 (depends on 5)

The method of claim 5, wherein said adjusting is used for training said foundation model.

### Claim 7 (depends on 6)

The method of claim 6, wherein training said AI engine is performed through a two-stage training process.

### Claim 8 (depends on 7)

The method of claim 7, wherein said two stage training process comprises a first training stage that is performed using an augmented dataset and a second training stage that is performed using said in-distribution dataset.

### Claim 9 (depends on 8)

The method of claim 8, wherein said contrastive learning model is used to maximize a cosine similarity between a plurality of embeddings of sentences of the two datasets from said same class.

### Claim 10 (depends on 9)

The method of claim 9, further comprising:
for each sentence in said out-of-distribution dataset, computing a cosine similarity with each sentence in said same class in said in-distribution dataset; and providing a score based on a level of similarity.

### Claim 11 (depends on 10)

The method of claim 10, further comprising comparing said scores indicating similarities based on similarity of each sentence in said out-of-distribution dataset.

### Claim 12 (depends on 11)

The method of claim 11, further comprising selecting a score with a highest similarity score amongst said out-of-distribution dataset.

### Claim 13 (depends on 12)

The method of claim 12, further comprising: selecting a sentence from said out-of-distribution dataset for classes with fewer than k sentences in said in-distribution dataset.

### Claim 14 (depends on 13)

The method of claim 13, further comprising appending selected sentences to a training dataset.

### Claim 15

A computer system for data augmentation for training an artificial intelligence (AI) engine, comprising:
one or more processors, one or more computer-readable memories and one or more computer-readable storage media;
program instructions, stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to encode an in-distribution dataset and an out-of-distribution dataset using a foundation model; wherein said encoding is done using a foundation model;
program instructions, stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to pair an in-distribution component with an out-of-distribution component in a same class using a contrastive model to provide a first set of paired components;
program instructions, stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to pair another in-distribution component with another out-of-distribution component in a different class using said contrastive learning model to provide a second set of paired components;
program instructions, stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to augment said first and second set of paired components to generate an augmented training dataset; and
program instructions, stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to adjust said foundation model by using said augmented training dataset to train said AI engine.

### Claim 16 (depends on 15)

The computer system of claim 15, wherein said adjusting comprises tuning and reiteratively fine-tuning said foundation model.

### Claim 17 (depends on 16)

The computer system of claim 16, wherein said in-distribution dataset is a labeled training dataset designated for a target task.

### Claim 18 (depends on 17)

The computer system of claim 17, wherein said another in-distribution component or said another out-of-distribution used to provide said second set of paired component may be the same as said in distribution component or said out-of-distribution component used to provide said first set of paired components.

### Claim 19

A computer program product for data augmentation used for training an artificial intelligence (AI) engine, comprising:
one or more computer readable storage media;
program instructions, stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to encode an in-distribution dataset and an out-of-distribution dataset using a foundation model; wherein said encoding is done using a foundation model;
program instructions, stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to pair an in-distribution component with an out-of-distribution component in a same class using a contrastive model to provide a first set of paired components;
program instructions, stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to pair another in-distribution component with another out-of-distribution component in a different class using said contrastive learning model to provide a second set of paired components;
program instructions, stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to augment said first and second set of paired components to generate an augmented training dataset; and
program instructions, stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to adjust said foundation model by using said augmented training dataset to train said AI engine.

### Claim 20 (depends on 19)

The computer program product of claim 19, wherein said adjusting comprises tuning and fine turning said foundation models.
