Learning Model Generator for Wireless Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection methods in wireless communication systems using machine learning face challenges with false negatives and missed detections due to reliance on unsupervised learning with large datasets, leading to oversized models and memory requirements, necessitating costly computing resources.
Innovation Solution
A learning model generator that divides feature data into groups, using a subset of data for training, including feature data from a primary group and a portion from other groups, to reduce model size and memory needs while maintaining accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large amount of training data is used to reduce false negative and missed detection, then detection accuracy is improved, but the size of the learning model and memory requirements increase
Solution Approach 1:
The patent divides the training data into multiple groups (first group, second group, third group) and trains multiple learning models (first learning model, second learning model, third learning model) separately on different groups. This segmentation allows the system to process data in manageable portions rather than requiring all data to be loaded into memory simultaneously, thereby reducing peak memory requirements while maintaining comprehensive coverage for accurate detection.
Solution Approach 2:
The patent uses only a portion of the available training data for each learning model (e.g., first learning model uses only the first group, second learning model uses second and third groups). This partial action approach reduces the amount of data that needs to be stored and processed at once, decreasing model size and memory requirements while still achieving reliable detection through ensemble voting.
2Reliability
If a large amount of training data is used to reduce false negative and missed detection, then detection accuracy is improved, but the computational cost increases
Solution Approach 1:
By segmenting the training data into multiple groups and training separate learning models on each group, the computational workload is distributed across multiple smaller training processes rather than one large training process. This segmentation reduces the computational burden on any single system, lowering energy consumption and computational cost while maintaining detection accuracy through the ensemble approach.
Solution Approach 2:
Each learning model is trained on only a portion of the total training data (e.g., first learning model on first group only), which reduces the computational resources required for each individual training process. The partial action approach allows multiple models to be trained efficiently with smaller datasets, reducing overall computational cost while the voting mechanism ensures accurate detection.
3Reliability
If feature data from multiple groups is used for training, then detection reliability is improved, but the complexity of data management increases
Solution Approach 1:
The patent organizes training data into distinct groups (first group, second group, third group) and assigns specific groups to specific learning models (first learning model, second learning model, third learning model). This segmentation creates a clear, systematic data management structure where each model has designated data sources, reducing the complexity of managing large datasets while ensuring comprehensive coverage for reliable detection.
Data Source
AI summary
The learning model generator includes a data division unit which groups multiple feature data, each of which indicates a feature, and a learning model generation unit which generates a learning model using feature data belonging to a first group among multiple groups formed by the data division unit, or the feature data belonging to the first group and a part of feature data belonging to other groups, as training data.