Device Type Classification Using Metric Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional device type classification methods in computer networks, such as rule-based systems, struggle to keep pace with the growing diversity of IoT devices, resulting in up to 40% of devices being classified as 'unknown' due to static classification rules' limitations, especially in weakly supervised settings with limited training data.
Innovation Solution
A device classification service employs metric learning to construct a distance function using telemetry data and side information, clustering devices into types and associating labels with these clusters, thereby adapting to new device types without relying solely on hard-coded knowledge.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If rule-based classification is used, then classification speed is maintained, but classification accuracy deteriorates as device diversity grows
Solution Approach 1:
The patent replaces the mechanical rule-based classification system with a machine learning-based metric learning system. Instead of using static classification rules that require manual updates, the system learns distance metrics automatically from data to classify devices into types, enabling it to adapt to new device types without manual rule creation.
Solution Approach 2:
The system changes the parameter space by learning a custom distance metric that transforms the original feature space into a more discriminative space. This learned metric adapts to the characteristics of different device types, improving classification accuracy while maintaining the ability to handle diverse and new device types.
2Adaptability or versatility
If more classification rules are added to cover new device types, then classification coverage improves, but system complexity increases
Solution Approach 1:
The patent replaces the complex mechanical system of maintaining and updating classification rules with a data-driven metric learning system. The system automatically learns distance metrics from training data, eliminating the need for manual rule creation and simplification while improving coverage of diverse device types.
3Measurement precision
If metric learning is applied, then classification accuracy improves, but computational requirements increase
Solution Approach 1:
The patent applies partial metric learning by focusing on learning only the essential distance metrics needed for classification rather than comprehensive model training. This approach achieves improved classification accuracy while reducing computational overhead compared to full machine learning model training and deployment.
Data Source
AI summary
In one embodiment, a device classification service receives telemetry data indicative of behavioral characteristics of a plurality of devices in a network. The service obtains side information for the telemetry data. The service applies metric learning to the telemetry data and side information, to construct a distance function. The service uses the distance function to cluster the telemetry data into device clusters. The service associates a device type label with a particular device cluster.