AI/ML Training Input Screening for Compromised UE Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile and cellular networks utilizing AI/ML algorithms for network optimizations are vulnerable to compromised user equipment (UE) inputs, which can skew performance and affect energy saving and mobility strategies.
Innovation Solution
Implement anomaly detection during data pre-processing and statistical correlation of inputs from similar locations to identify and isolate compromised UEs, sharing malicious UE behavior information across network nodes, and using core network entities to manage compromised UE lists.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If AI/ML algorithms utilize data from various network entities including UEs for network optimizations, then network optimization performance is improved, but vulnerability to compromised UE inputs increases
Solution Approach 1:
The patent introduces a data validation intermediary layer between UE data sources and AI/ML algorithms. This intermediary validates incoming data from UEs using multiple criteria (data quality assessment, source reliability evaluation, anomaly detection) before feeding data to AI/ML models, thereby maintaining optimization performance while filtering out compromised inputs
Solution Approach 2:
The patent implements feedback mechanisms where AI/ML model outputs and network performance metrics are continuously monitored. When anomalies or performance degradation are detected, the system feeds back to adjust data collection strategies, revalidate data sources, or modify model parameters, creating a closed-loop system that adapts to potential compromises
2Measurement precision
If UEs are provisioned to actively contribute to AI/ML operations, then model training quality is improved, but detection of compromised UEs becomes more difficult
Solution Approach 1:
The patent employs data characterization techniques that assign different 'colors' or labels to data from different UE sources based on their reliability, quality metrics, and historical performance. This visual/metaphorical classification system makes it easier to identify and isolate compromised UEs by highlighting anomalies in data patterns, sources, or contributions
Solution Approach 2:
The patent segments the UE population into different groups based on trust levels, data quality, and contribution patterns. By dividing UEs into segments (e.g., high-trust, medium-trust, low-trust groups), the system can apply different validation strategies and detect compromised UEs more effectively without compromising overall model training quality
Data Source
Figure 1
Figure 2
Figure 3
AI summary
There are provided measures for improved robustness of artificial intelligence or machine learning capabilities against compromised input. Such measures exemplarily comprise receiving, from a first machine learning model training data collection entity, first machine learning model training input data, analyzing said first machine learning model training input data for malicious input detection, deducing, based on a result of said analyzing, whether said first machine learning model training data collection entity is suspected to be compromised, and transmitting, if said first machine learning model training data collection entity is suspected to be compromised, information to a core network entity indicative of that said first machine learning model training data collection entity is suspected to be compromised.