Network Switch Local Classifier Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing process of building or updating a classifier using labeled data samples imposes a significant burden on communications networks, requiring powerful server computers and increased network traffic, especially with large datasets.
Innovation Solution
Leveraging the processing resources of communications networks, specifically switches, to perform classification tasks, including learning and applying classifiers, by using training data packets with labeled fields to train and forward data packets based on forwarding information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training data is provided to central server computers for learning a classifier, then the classifier can be built or updated, but it imposes a large burden on communications networks and requires powerful server computers
Solution Approach 1:
The patent segments the classification task from the central server and distributes it to network switches. The switch learns a classifier locally using training data packets, separating the learning function from the central server. This allows the server to remain simple while switches perform classification independently, reducing network burden and server requirements.
Solution Approach 2:
The patent moves the classification capability from the computational dimension (central server) to the network infrastructure dimension (switches). By embedding learning functionality within standard network switches that already exist in the network, the system leverages unused processing resources at the network edge rather than relying on centralized computational power.
2Measurement precision
If training data is sent through the communications network to server computers, then the classifier can be trained, but it imposes a large burden on network traffic
Solution Approach 1:
The patent segments training data into specialized training data packets that are processed locally by switches. Instead of sending all training data to central servers, the data is distributed to switches that need it, and switches perform local learning. This segmentation reduces redundant data transmission and network bandwidth consumption.
Solution Approach 2:
The switch performs classification learning independently using received training data packets, without requiring continuous communication with the central server. The switch serves its own classification needs by learning locally, reducing the need for ongoing network communication for model updates and minimizing bandwidth usage.
Data Source
AI summary
A switch receives a data packet containing training information. The switch learns a classifier based on the training information in the data packet, the classifier useable to classify data into at least one category.