Data Classification Using Shared Features for Unknown Families

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing machine learning methods struggle to accurately classify data from unknown families in cyber security, especially when features change drastically or labeled data is unavailable.

Innovation Solution

A data classification device that extracts features shared between subclasses within a class, causing classification into subclasses to fail, and uses these features to classify new data into a class.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning uses features specific to source data for classification, then classification accuracy for known families improves, but classification accuracy for unknown families deteriorates

Engineering Contradiction:

Improveclassification accuracyVSAvoidability to classify unknown families

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent segments features into two categories: shared features (common across multiple families) and specific features (unique to individual families). The system selectively uses shared features for classifying unknown families while using specific features for known families, thereby resolving the contradiction between accuracy for known families and adaptability to unknown families.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent changes the parameter selection based on whether the data family is known or unknown. For unknown families, it switches to using shared features instead of specific features, effectively changing the classification parameters to adapt to the unknown nature of the data, thus improving both accuracy and adaptability.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If machine learning extracts features that distinguish subclasses within a class, then subclass classification accuracy improves, but class-level classification robustness deteriorates

Engineering Contradiction:

Improvesubclass classification accuracyVSAvoidclass-level classification robustness

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The patent extracts and separates specific features that are unique to subclasses from the overall feature set. By taking out these subclass-specific features, the system can use them for accurate subclass classification while relying on shared features for robust class-level classification, thus resolving the contradiction between subclass accuracy and class-level robustness.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces a hierarchical dimension to feature usage: shared features operate at the class level while specific features operate at the subclass level. This dimensional separation allows the system to achieve both subclass classification accuracy and class-level classification robustness by operating in different feature dimensions simultaneously.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS12468804B2Data classification device, data classification method, and data classification program

Publication Date: 2025.11.11 NT T INC

AI summary

A data classification device includes: a known data input unit that receives an input of known data, the known data being data already classified into a class and a subclass subordinate to the class; a feature extraction unit that extracts, from features included in the known data, a feature that causes classification of the known data belonging to the same class into a subclass using the feature to fail; and a classification unit that classifies classification target data into a class using the feature extracted by the feature extraction unit.