Stable Device Clustering via Autoencoder Telemetry Analysis

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

In IoT networks, the increasing diversity of devices makes it challenging to accurately identify device types due to the lack of ground truth, leading to instability in device clusters and potential misclassification, especially with the proliferation of non-traditional devices.

Innovation Solution

A device classification service that uses telemetry data to repeatedly assign devices to clusters, determining stability loss and obtaining device type labels once cluster assignments stabilize, leveraging a combination of reconstruction and classification losses to optimize feature representation for clustering, and employing autoencoders to learn stable representations of devices.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If clustering is applied to device telemetry data to identify device types, then device classification can be performed without ground truth, but cluster instability leads to misclassification

Engineering Contradiction:

Improvedevice classification capabilityVSAvoidcluster stability

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent implements dynamic cluster assignment by repeatedly assigning devices to clusters over multiple iterations and using stability metrics to determine when clusters have converged. The system dynamically adjusts cluster assignments based on telemetry data patterns while maintaining stability through repeated measurements, allowing the classification system to adapt to device behavior patterns without sacrificing reliability

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent employs feedback mechanisms by calculating stability metrics based on previous cluster assignments and using this feedback to refine subsequent cluster assignments. The system monitors whether devices remain in the same cluster across multiple iterations and uses this feedback to determine when stable classification has been achieved, preventing misclassification due to transient clustering patterns

Inventive Principle:

Principle #23Feedback

2Adaptability or versatility

If more non-traditional IoT devices are connected to the network, then network service diversity increases, but device type identification becomes more challenging

Engineering Contradiction:

Improvenetwork service diversityVSAvoiddevice type identification

Core Design Contradiction:

Adaptability or versatilityVSDifficulty of detecting and measuring

Solution Approach 1:

The patent implements self-service classification by enabling devices to be automatically classified based on their own telemetry data patterns without requiring manual identification or ground truth labels. The clustering algorithm analyzes behavioral patterns, traffic characteristics, and operational metrics inherent to each device type, allowing the system to self-organize and identify device types autonomously as new IoT devices connect to the network

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent utilizes parameter changes in device telemetry data to identify device types by monitoring variations in communication patterns, data transmission intervals, protocol usage, and operational parameters. The system detects device types by analyzing how these parameters change over time and comparing patterns across multiple devices, enabling identification of diverse IoT device types through their characteristic parameter profiles

Inventive Principle:

Principle #35Parameter changes

3Productivity

If cluster assignments are made based on initial telemetry data, then device classification is obtained quickly, but misclassifications occur due to insufficient stabilization

Engineering Contradiction:

Improveclassification speedVSAvoidclassification accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent applies preliminary action by performing multiple preliminary cluster assignments and stability checks before finalizing device classifications. Rather than making a single hurried classification decision, the system repeatedly assigns devices to clusters, checks for stability, and only finalizes classifications when stability criteria are met, ensuring both speed and accuracy through structured preliminary evaluation

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent implements periodic action by conducting cluster assignments and stability assessments at regular intervals rather than continuously or only once. The system periodically re-evaluates cluster assignments, checks stability metrics, and updates classifications when changes are detected, maintaining both efficient processing throughput and high classification accuracy through rhythmic evaluation cycles

Inventive Principle:

Principle #19Periodic action

Data Source

PatentUS11451456B2Learning stable representations of devices for clustering-based device classification systems

Publication Date: 2022.09.20 CISCO TECHNOLOGY INC

AI summary

In one embodiment, a device classification service obtains telemetry data for a plurality of devices in a network. The device classification service repeatedly assigns the devices to device clusters by applying clustering to the obtained telemetry data. The device classification service determines a measure of stability loss associated with the cluster assignments. The measure of stability loss is based in part on whether a device is repeatedly assigned to the same device cluster. The device classification service determines, based on the measure of stability loss, that the cluster assignments have stabilized. The device classification service obtains device type labels for the device clusters, after determining that the cluster assignments have stabilized.