Network Traffic Classifier Adaptation via Parameter Sharing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network traffic classification systems face challenges in adapting to specific network conditions due to privacy concerns and the need for robust training data, as entities are hesitant to share sensitive data, leading to classifiers that may not accurately identify malicious traffic flows.
Innovation Solution
A method where a generic classifier is trained using centrally-curated data and then adapted using distinct private training data sets within the target network, ensuring privacy by only sharing classifier parameters, allowing for fine-tuning based on local conditions without exposing sensitive information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If entities share sensitive training data centrally, then classifier robustness improves, but data privacy is compromised
Solution Approach 1:
The training data remains segmented and distributed across multiple private entities rather than being centralized. Each entity maintains its own training data locally, and only trained model parameters (not the raw data) are shared with the central server for aggregation. This segmentation preserves data privacy while still enabling collaborative model improvement through parameter sharing.
Solution Approach 2:
The patent introduces a central server as an intermediary that facilitates model parameter aggregation without directly handling or storing sensitive training data. The server receives trained parameters from various entities, aggregates them to create an improved global model, and redistributes the updated parameters back to entities. This intermediary mechanism enables collaboration while maintaining data privacy boundaries.
2Measurement precision
If generic classifier is trained on diverse network data, then classification accuracy improves, but adaptability to specific network conditions deteriorates
Solution Approach 1:
The classifier model is made dynamic through iterative retraining and parameter updates. Entities can retrain local models using their own network-specific data and send updated parameters to the central server. This allows the global model to dynamically adapt to different network conditions while maintaining the benefits of diverse training data, achieving both accuracy and adaptability.
Solution Approach 2:
The patent enables local customization of classifier parameters at each entity based on their specific network conditions. While the global model provides general classification accuracy from diverse data, each entity can fine-tune local parameters to match their specific network environment, achieving both general accuracy and local adaptability simultaneously.
Data Source
AI summary
In one embodiment, a device in a first network receives traffic flow information regarding a plurality of traffic flows in the first network. The device labels the traffic flow information by associating classifier labels to the traffic flow information. The device receives a generic traffic classifier that was trained using a training data set that comprises labeled traffic flow information for a plurality of other networks and excludes the traffic flow information regarding the plurality of traffic flows in the first network. The device acclimates the generic traffic classifier to the first network using the labeled traffic flow information regarding the plurality of traffic flows in the first network.