Embedded Neural Network Matrix for Fast Wireless Anomaly Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems face challenges in efficiently troubleshooting network anomalies in wireless networks due to resource constraints at base stations, leading to prolonged inference generation and processing times in central or edge clouds, which hinder real-time anomaly detection and resolution.
Innovation Solution
Embedding neural networks as a matrix within network devices using a network server to determine target KPIs, optimize resource usage, and predict future anomalies, thereby reducing processing time and resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ML model is deployed on central or edge cloud, then prediction accuracy is maintained, but processing time increases
Solution Approach 1:
The patent segments the ML model into two parts: training phase on central/edge cloud and inference phase on base station. The model is trained centrally using comprehensive data, then the trained model is deployed to base stations for local inference, dividing the workload to achieve both accuracy and speed
Solution Approach 2:
The patent performs preliminary model training on the central or edge cloud before deployment. The ML model is pre-trained using historical and real-time KPI data, so that when deployed to base stations, the inference can be executed immediately without requiring complex training computations at the edge
2Loss of time
If ML model is implemented in base stations, then processing time is reduced, but resource constraints prevent implementation
Solution Approach 1:
The patent creates a copy of the trained ML model and deploys it to base stations. Instead of implementing the full training capability in resource-constrained base stations, the trained model parameters and structure are copied to the base station, enabling local inference with minimal computational resources
Solution Approach 2:
The patent uses lightweight model representations that can be deployed to base stations with limited resources. The model is optimized for edge deployment with reduced computational requirements, allowing inference on devices with constrained CPU, memory, and energy resources
3Measurement precision
If inference is generated in central cloud, then model accuracy is maintained, but troubleshooting time increases
Solution Approach 1:
The patent segments the intelligence into model training (centralized) and model execution (distributed). The central cloud handles the complex task of model training and updates, while base stations handle real-time inference, enabling faster local troubleshooting while maintaining centralized model quality control
Solution Approach 2:
The patent introduces an intermediary deployment mechanism where the trained model is transferred from central cloud to base stations. This intermediary step allows the system to maintain the accuracy benefits of centralized training while achieving the speed benefits of distributed inference
Data Source
AI summary
Embodiments herein disclose a method and a device for embedding neural networks as a matrix for a network device in wireless networks. The method includes receiving s from the network device. Further, the method also includes determining the KPI among the plurality of KPIs as target KPIs that related to a network anomaly using a ML model. Further, the method also includes determining a correlation of the target KPI with the plurality of KPIs for the network anomaly using the ML model. Further, the method also includes determining the matrix indicating a relation of the target KPI with the plurality of KPIs. Furthermore, the method includes optimizing a resource of the network device by embedding the matrix in the network device.