Privacy-Aware Model Generation for Hybrid ML Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current network assurance systems face challenges in effectively analyzing network performance using machine learning models, particularly in hybrid systems where privacy concerns restrict the sharing of telemetry data, leading to suboptimal model performance due to limited data availability and computational resources.
Innovation Solution
A privacy-aware approach is introduced, where a network assurance service clusters measurements from local devices, computes aggregated metrics, and shares these with a remote service to train machine learning models using synthetic datasets combining local and remote network data, ensuring differential privacy and enhancing model performance without compromising user privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If telemetry data is shared with remote service for model training, then model performance is improved, but user privacy is compromised
Solution Approach 1:
The patent extracts only the essential aggregated metrics from the original telemetry data before sharing with the remote service. The network assurance service processes raw measurements locally, computes aggregated metrics that capture network performance characteristics, and transmits only these processed metrics to the remote service for model training, leaving the detailed raw data local to protect privacy.
Solution Approach 2:
The patent introduces an intermediary processing layer (the network assurance service) that stands between the raw telemetry data and the remote service. This intermediary aggregates and processes data locally, transforming sensitive raw measurements into anonymized aggregated metrics that can be safely shared while preserving the essential information needed for model training.
2Object-affected harmful factors
If all raw telemetry data is kept local for privacy, then user privacy is protected, but model performance deteriorates due to limited data availability
Solution Approach 1:
The patent transforms the data representation from detailed raw measurements to aggregated metrics through local processing. By changing the parameter representation (from individual device measurements to cluster-level aggregations), the system enables remote model training with sufficient data while maintaining privacy protection through the aggregation transformation.
3Object-affected harmful factors
If aggregated metrics are shared instead of raw data, then user privacy is protected, but data quality for training may be reduced
Solution Approach 1:
The patent applies partial action by sharing only the necessary aggregated metrics rather than all raw data. The aggregation process retains the essential performance characteristics needed for model training while discarding identifying details, achieving a balance between privacy protection and data utility for training purposes.
Data Source
AI summary
In one embodiment, a network assurance service executing in a local network clusters measurements obtained from the local network regarding a plurality of devices in the local network into measurement clusters. The network assurance service computes aggregated metrics for each of the measurement clusters. The network assurance service sends a machine learning model computation request to a remote service outside of the local network that includes the aggregated metrics for each of the measurement clusters. The remote service uses the aggregated metrics to train a machine learning-based model to analyze the local network. The network assurance service receives the trained machine learning-based model to analyze performance of the local network. The network assurance service uses the receive machine learning-based model to analyze performance of the local network.