Filtering Abnormal Network Parameters for Prediction Model Training

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing communication networks face challenges in ensuring that network parameter values used for training prediction models are standard and expected, as malicious or misconfigured devices can introduce erroneous values, leading to inaccurate predictions.

Innovation Solution

A method and apparatus for filtering abnormal network parameter values in a communication network by determining, based on an expected correlation with trusted values, whether received values are anomalies, and only using non-anomalous values for training a prediction model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If machine learning techniques are used to train a network prediction model with continuous network parameter values, then the network performance and traffic optimization are enhanced, but the prediction model becomes vulnerable to erroneous or corrupted values from malicious or misconfigured devices

Engineering Contradiction:

Improvenetwork performance optimizationVSAvoidprediction model accuracy

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent applies preliminary action by implementing anomaly detection and filtering mechanisms before the training process. The system pre-establishes expected parameter ranges and correlation patterns, then proactively identifies and removes anomalous values from the training data stream, preventing corrupted data from affecting model accuracy while maintaining continuous learning capabilities.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary filtering layer between data collection and model training. This intermediary component validates network parameter values against expected patterns and correlations, acting as a gatekeeper that allows only legitimate data to reach the training model, thus decoupling the model from direct exposure to malicious or misconfigured device data.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Ease of manufacture

If all network parameter values are used for training the prediction model, then the model training is simplified and continuous, but corrupted values from malicious devices can be included and used to update the model

Engineering Contradiction:

Improvetraining process simplicityVSAvoiderroneous or corrupted values

Core Design Contradiction:

Ease of manufactureVSObject-affected harmful factors

Solution Approach 1:

The patent extracts and removes anomalous values from the training data stream before they can reach the model. By identifying and separating corrupted values based on deviation from expected patterns and correlations, the system maintains a clean training dataset while preserving the continuous training process, thus eliminating harmful factors without complicating the overall training methodology.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent converts the potential harm of corrupted data into a benefit by using anomaly detection to identify and flag erroneous values. The system leverages the presence of corrupted data to strengthen its filtering mechanisms, turning what would be a contaminant into a signal for improving data validation capabilities and enhancing overall system robustness.

Inventive Principle:

Principle #22Blessing in disguise (Convert harm into benefit)

3Reliability

If network parameter values from untrusted devices are filtered based on anomaly detection, then the prediction model accuracy is maintained, but the device complexity increases due to additional filtering mechanisms

Engineering Contradiction:

Improveprediction model accuracyVSAvoidfiltering mechanism complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies parameter changes by transforming the filtering approach from complex structural modifications to parameter-based validation. Instead of implementing complex filtering algorithms, the system changes the parameters used for data validation by establishing expected ranges and correlation patterns, making the filtering mechanism simpler and more efficient while maintaining high prediction accuracy.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12316508B2Method and system for filtering of abnormal network parameter values prior to being used in training of a prediction model in a communication network

Publication Date: 2025.05.27 TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)

AI summary

A method and system of filtering abnormal network parameter values are described. A network node receives first values of first network parameters that are not trusted by the network node. The network node determines, based on an expected correlation between the first values of the first network parameters and second values of second network parameters, whether the first values of the first network parameters are anomalies. Responsive to determining that the first values of the first network parameters are not anomalies, the network node causes the first values to be used as input to train a network prediction model that is used for optimization of traffic in the network, and responsive to determining that the first values of the first network parameters are anomalies, the network node causes the first values to not be used to train the network prediction model.