Kernel Herding Representative Point Extraction for IoT Anomaly Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Anomaly detection systems for IoT devices face challenges in ascertaining normal communication features due to the diverse nature of IoT device communications, which involves various protocols and characteristics, making it difficult to determine traffic feature amount tendencies through statistical processing alone.

Innovation Solution

The use of kernel herding as a method to extract representative points from traffic feature amounts for learning, allowing the system to provide data for ascertaining the tendency of these features, and when combined with a generative model like VAE, helps in understanding what kind of communication is considered normal by the model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If statistical processing such as average value or median value calculation is performed on traffic feature amounts, then processing simplicity is improved, but the ability to ascertain communication tendencies deteriorates due to the diverse nature of IoT device communications

Engineering Contradiction:
Improveprocessing simplicityVSAvoidcommunication tendency information
Core Design Contradiction:
Ease of manufactureVSLoss of information

Solution Approach 1:

The patent introduces an intermediary component (tendency ascertaining unit) that processes traffic feature amounts to extract meaningful communication tendencies. This intermediary transforms raw diverse traffic data into structured tendency information that can be used by the anomaly detection system, resolving the contradiction between processing simplicity and information preservation.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If the anomaly detection system detects anomalies without knowing all malignant conditions, then detection coverage is improved, but the ability to understand detection reasons deteriorates

Engineering Contradiction:
Improvedetection coverageVSAvoiddetection reason information
Core Design Contradiction:
Adaptability or versatilityVSLoss of information

Solution Approach 1:

The patent implements a feedback mechanism where the tendency ascertaining unit provides contextual information about normal communication patterns back to the anomaly detection system. This feedback loop enables the system to not only detect anomalies but also understand the reasons for detection by comparing detected traffic against established communication tendencies.

Inventive Principle:
Principle #23Feedback

3Reliability

If generative models like VAE are used to learn normal communication patterns, then anomaly detection capability is improved, but the complexity of ascertaining what constitutes normal communication deteriorates

Engineering Contradiction:
Improveanomaly detection capabilityVSAvoidmodel interpretation complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts interpretable tendency information from the complex generative model outputs through the tendency ascertaining unit. This extraction process separates the black-box model predictions from the meaningful communication patterns, allowing operators to understand what constitutes normal communication without being overwhelmed by model complexity.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4092581B1Learning device, learning method, and learning program
Publication Date: 2025.04.02 NT T INC
  • EP4092581B1 patent drawingFigure 1
  • EP4092581B1 patent drawingFigure 2
  • EP4092581B1 patent drawingFigure 3

AI summary

A learning device (10) includes an acquisition unit (131) that acquires pieces of communication data for learning, a feature amount extraction unit (132) that extracts feature amounts of the communication data, a training unit (1331) that trains a VAE model (121) with the feature amounts of the communication data, a representative point extraction unit (1332) that extracts representative points of the feature amounts of the communication data using kernel herding, and an output unit that outputs the representative points extracted by the representative point extraction unit (1332).