Core Concepts
A novel federated transfer learning method, Federated Transfer Component Analysis (FTCA), is proposed to efficiently transfer profiling knowledge between different Virtual Network Function (VNF) types while preserving data privacy.
Abstract
The paper proposes a Federated Transfer Component Analysis (FTCA) method to address the challenges of VNF profiling, which is time-consuming and requires collecting profiling data from all VNF types. FTCA leverages federated learning and transfer learning to transfer the profiling knowledge from a well-profiled source VNF to a less-profiled target VNF, while keeping the raw profiling data private.
The key steps of FTCA are:
- The source VNF trains a Generative Adversarial Network (GAN) model on its profiling data and sends the trained model to the target VNF.
- The target VNF uses the received GAN model to generate synthetic source VNF data and combines it with its own limited profiling data.
- The target VNF then applies transfer component analysis (TCA) to map the combined data into a new feature space where the difference between the source and target domains is minimized.
- Finally, regression models are trained on the transformed data to predict the resource configurations (CPU, memory, link capacity) of the target VNF.
Experiments on three VNF types (SNORT Inline, SNORT Passive, and virtual firewall) show that FTCA can effectively predict the target VNF resource configurations, reducing the RMSE by up to 38.5% and improving the R-squared metric up to 68.6% compared to directly training on the target VNF data.
Stats
The RMSE index of the regression model decreases by 38.5% and the R-squared metric advances up to 68.6%.
Quotes
"Combining two learning methods, federated transfer component analysis (FTCA) is proposed in this paper as a novel method under federated transfer learning (FTL)."
"Experiments show that the proposed FTCA can effectively predict the required resources for the target VNF."