Semantic Data Augmentation for Automated Feature Engineering

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current feature engineering in machine learning is predominantly manual, limiting the efficiency and scalability of data augmentation processes.

Innovation Solution

Automated data augmentation through semantic transforms, where variables are mapped to concepts, and suitable transforms are selected and applied to generate expressions, enhancing data sets for machine learning applications.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If manual feature engineering is used, then data augmentation can be performed, but the process is time-consuming and lacks scalability

Engineering Contradiction:

Improvedata augmentation efficiencyVSAvoidmanual feature engineering time

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The system performs self-service by automatically mapping variables to concepts and selecting appropriate semantic transforms without human intervention. The automated pipeline independently completes feature engineering tasks that previously required manual analyst effort, thereby eliminating time loss while maintaining data augmentation quality

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system performs preliminary action by pre-defining a comprehensive library of semantic transforms and concepts before the actual feature engineering process. This pre-prepared knowledge base enables rapid automatic transformation of variables during data augmentation, significantly improving productivity without requiring manual analysis at execution time

Inventive Principle:

Principle #10Preliminary action

2Productivity

If automated data augmentation is implemented, then productivity increases, but the complexity of the system increases

Engineering Contradiction:

Improvefeature engineering throughputVSAvoidautomated pipeline complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The system uses an intermediary approach by introducing a structured conceptual layer between raw variables and final features. The variable-to-concept mapping acts as an intermediary step that simplifies the overall process, making the automated pipeline more manageable despite increased complexity through standardized intermediate representations

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The automated feature engineering process is segmented into distinct modular steps: variable mapping to concepts, concept matching with semantic transforms, transform selection, and expression generation. This segmentation reduces system complexity by breaking down the monolithic automated pipeline into independent, manageable modules that can be developed and maintained separately

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12561613B2Data augmentation using semantic transforms

Publication Date: 2026.02.24 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

A method of data augmentation includes receiving, by a processor, a set of data including a plurality of variables, mapping each variable to one or more target concepts associated with a name of each variable, and acquiring a set of semantic transforms, each semantic transform including a function applied to one or more concepts mapped to a respective variable. The method also includes comparing the one or more target concepts to the one or more concepts of each semantic transform, selecting at least one semantic transform based on the comparing, generating an expression for each selected semantic transform, each expression configured to apply a function of a selected semantic transform to at least one of the plurality of variables, and augmenting the set of data for use in an application by adding each expression to the set of data.