Abstract:The generative adversarial networks (GANs) have recently been applied to estimating the distribution of independent and identically distributed data, and have attracted a lot of research attention. In this paper, we use the blocking technique to demonstrate the effectiveness of GANs for estimating the distribution of stationary time series. Theoretically, we derive a non-asymptotic error bound for the Deep Neural Network (DNN)-based GANs estimator for the stationary distribution of the time series. Based on our theoretical analysis, we propose an algorithm for estimating the change point in time series distribution. The two main results are verified by two Monte Carlo experiments respectively, one is to estimate the joint stationary distribution of $5$-tuple samples of a 20 dimensional AR(3) model, the other is about estimating the change point at the combination of two different stationary time series. A real world empirical application to the human activity recognition dataset highlights the potential of the proposed methods.

What problem does this paper attempt to address?

This paper attempts to solve two main problems: 1. **Estimation of the joint distribution of time series**: - The main objective of the paper is to verify the effectiveness of generative adversarial networks (GANs) based on deep neural networks (DNN) in estimating the joint distribution of stationary time series. Specifically, the paper aims to estimate the joint stationary distribution $\pi_l$ of time series, that is, the joint distribution of sample groups of length $l$. - Through the blocking technique, the author proves that GANs based on DNN can effectively learn the joint distribution of time series and derives non - asymptotic error bounds. This result shows that the GANs generator can stably generate samples very close to the real data distribution. 2. **Change - point estimation in time series**: - In many practical applications, time series may experience structural changes, for example, the distribution may change suddenly. Therefore, estimating these change - points becomes very important. - Based on the above theoretical results, the paper proposes an algorithm based on GANs to estimate a single change - point in time series. This algorithm identifies the location of the change - point by comparing the distribution differences in different time periods. ### Specific problem description - **Joint distribution estimation**: For the stationary time series $\{X_n\}_{n\in\mathbb{N}}$, its joint stationary distribution $\pi_l$ refers to the joint distribution of sample groups of length $l$. The paper estimates this joint distribution by using GANs and proves its effectiveness through non - asymptotic error bounds. - **Change - point estimation**: When the distribution of a time series changes at a certain time point $\tau$, how to accurately estimate the location of this change - point is an important research topic. The paper proposes a method based on GANs to identify change - points by comparing the distribution differences in different time periods. ### Theoretical contributions - **Non - asymptotic error bounds**: The paper derives a non - asymptotic error bound to measure the gap between the joint distribution estimated by the GANs generator and the real joint distribution. This error bound shows that as the sample size increases, the estimation accuracy of the GANs generator will gradually improve. - **Change - point estimation algorithm**: Based on the above theoretical results, the paper proposes an effective change - point estimation algorithm. This algorithm gradually narrows the possible location range of the change - point through the blocking technique and the calculation of the loss function, and finally determines the specific location of the change - point. ### Experimental verification - The paper verifies its theoretical results through two Monte Carlo experiments. The first experiment shows that GANs can successfully generate high - dimensional time series samples with similar correlations to the real data. The second experiment shows the effectiveness of the change - point estimation algorithm in different combinations of stationary time series. - In addition, the paper also shows the potential of the proposed algorithm in practical applications through an actual human activity recognition data set. In conclusion, through theoretical analysis and experimental verification, this paper proves the effectiveness and potential application value of GANs in the estimation of the joint distribution of time series and change - point estimation.

Distribution estimation and change-point estimation for time series via DNN-based GANs

Neural Stochastic Differential Equations with Change Points: A Generative Adversarial Approach

Dual Distribution Matching GAN

Directed Chain Generative Adversarial Networks

Distributionally robust chance constrained programming with generative adversarial networks (GANs)

Generative adversarial networks in time series: A survey and taxonomy

Conditional Sig-Wasserstein GANs for Time Series Generation

DEGAN: Time Series Anomaly Detection using Generative Adversarial Network Discriminators and Density Estimation

A New Distributed Method for Training Generative Adversarial Networks

Automatic Change-Point Detection in Time Series via Deep Learning

Generative Adversarial Networks in Time Series: A Systematic Literature Review

Time series data augmentation method of small sample based on optimized generative adversarial network

A data imputation method for multivariate time series based on generative adversarial network.

Time Series Simulation by Conditional Generative Adversarial Net

D-GAN: Deep Generative Adversarial Nets for Spatio-Temporal Prediction

DLT-GAN: Dual-Layer Transfer Generative Adversarial Network-Based Time Series Data Augmentation Method

If You Like It, GAN It. Probabilistic Multivariate Times Series Forecast With GAN

TTS-GAN: A Transformer-based Time-Series Generative Adversarial Network

Time-series Generative Adversarial Networks

Quant GANs: deep generation of financial time series