[Paper Review] Generative adversarial networks in time series: A survey and taxonomy

Eoin Brophy, Zhengwei Wang+2 more|arXiv (Cornell University)|Jul 23, 2021

Generative Adversarial Networks and Image Synthesis96 references46 citations

TL;DR

This paper surveys GAN variants for time series, proposing a taxonomy that splits discrete-variant and continuous-variant GANs, and discusses architectures, evaluation metrics, and privacy considerations.

ABSTRACT

Generative adversarial networks (GANs) studies have grown exponentially in the past few years. Their impact has been seen mainly in the computer vision field with realistic image and video manipulation, especially generation, making significant advancements. While these computer vision advances have garnered much attention, GAN applications have diversified across disciplines such as time series and sequence generation. As a relatively new niche for GANs, fieldwork is ongoing to develop high quality, diverse and private time series data. In this paper, we review GAN variants designed for time series related applications. We propose a taxonomy of discrete-variant GANs and continuous-variant GANs, in which GANs deal with discrete time series and continuous time series data. Here we showcase the latest and most popular literature in this field; their architectures, results, and applications. We also provide a list of the most popular evaluation metrics and their suitability across applications. Also presented is a discussion of privacy measures for these GANs and further protections and directions for dealing with sensitive data. We aim to frame clearly and concisely the latest and state-of-the-art research in this area and their applications to real-world technologies.

Motivation & Objective

• Provide a comprehensive review of GANs applied to time series generation and forecasting.
• Propose a taxonomy distinguishing discrete-variant and continuous-variant time series GANs.
• Summarize architectures, loss functions, evaluation metrics, datasets, and privacy considerations in this domain.

Proposed method

• Present a taxonomy of time series GANs into discrete-variant and continuous-variant categories.
• Summarize and compare representative models (e.g., SeqGAN, C-RNN-GAN, RCGAN, TimeGAN, SigCWGAN) and their architectures.
• Discuss training challenges (stability, vanishing gradients, mode collapse), evaluation approaches, and privacy risks.
• Review popular datasets used for time series GAN research and the lack of standard benchmarking datasets.
• Describe privacy-preserving aspects and regulatory considerations related to synthetic time series data.

Experimental results

Research questions

• RQ1 What are the main types of time series GANs (discrete-variant vs. continuous-variant) and their characteristic architectures?
• RQ2 What are the common training challenges and evaluation metrics for time series GANs?
• RQ3 What privacy risks and protections are relevant for synthetic time series data?
• RQ4 What datasets and benchmarks are used in time series GAN research, and what gaps exist in standardization?

Key findings

• Introduces a two-part taxonomy: discrete-variant and continuous-variant time series GANs.
• Synthesizes architectures, losses, and applications across leading time series GAN models.
• Highlights training stability, evaluation, and privacy as core challenges for time series GANs.
• Identifies lack of standardized benchmarking datasets and metrics for time series GANs.
• Documents popular datasets and real-world applications reviewed in the literature.
• Reviews privacy-preserving approaches and regulatory considerations for synthetic data.