Doppelganger: Synthetic Networked Time Series Generation via GANs

Developed by fjxmlzn for the IMC 2020 conference, this framework generates synthetic networked time series data using Generative Adversarial Networks (GANs). It provides a methodology for creating privacy-preserving versions of sensitive network measurement data while maintaining temporal fidelity.

Use Cases

• Generating synthetic network traffic traces to facilitate privacy-preserving data sharing
• Evaluating GAN fidelity using temporal correlation metrics on networked data
• Training time-series forecasting models when real-world data is restricted by privacy policies

Strengths

• Recognized as an IMC 2020 Best Paper Finalist for its technical approach
• Specifically optimized for high-dimensional networked time series
• Released under the permissive BSD-3-Clause-Clear license

Limitations

• The repository is a generation framework, meaning data quality depends on the user's training parameters
• Requires significant GPU resources for GAN training and synthesis
• Lack of pre-packaged large-scale datasets in the metadata

Provenance

Source

IMC 2020 publication 'Using GANs for Sharing Networked Time Series Data'

Collection Method

Synthetic generation via Generative Adversarial Networks

Freshness

Last updated November 2023; based on 2020 research.