Abstract:In this paper, we focus on the generation of time-series data using neural networks. It is often the case that input time-series data have only one realized (and usually irregularly sampled) path, which makes it difficult to extract time-series characteristics, and its noise structure is more complicated than i.i.d. type. Time series data, especially from hydrology, telecommunications, economics, and finance, exhibit long-term memory also called long-range dependency (LRD). The main purpose of this paper is to artificially generate time series with the help of neural networks, making the LRD of paths into account. We propose fSDE-Net: neural fractional Stochastic Differential Equation Network. It generalizes the neural stochastic differential equation model by using fractional Brownian motion with a Hurst index larger than half, which exhibits the LRD property. We derive the solver of fSDE-Net and theoretically analyze the existence and uniqueness of the solution to fSDE-Net. Our experiments with artificial and real time-series data demonstrate that the fSDE-Net model can replicate distributional properties well.

What problem does this paper attempt to address?

### The Problem the Paper Attempts to Solve This paper aims to address the problem of generating time series data with Long-term Memory (LRM). Specifically, the paper focuses on how to use neural networks to generate time series data, especially when the input time series data typically has only one realization path and irregular sampling, making it difficult to extract time series features, and its noise structure is more complex than the independent and identically distributed (i.i.d.) type. Time series data, especially in fields such as hydrology, telecommunications, economics, and finance, exhibit long-term memory characteristics, also known as Long Range Dependency (LRD). Therefore, the main goal of the paper is to artificially generate time series data through neural networks, taking into account the LRD characteristics of the path. ### Solution To solve the above problem, the paper proposes fSDE-Net (Neural Fractional Stochastic Differential Equation Network), which extends the traditional neural stochastic differential equation model by using fractional Brownian motion (fBm) with a Hurst index greater than 0.5. fBm can exhibit LRD characteristics, thereby better simulating the complex noise structure in actual time series data. ### Main Contributions 1. **Proposing fSDE-Net**: By extending the existing SDE-Net, using fBm to model noise, and theoretically proving the existence and uniqueness of the fSDE solution. 2. **Constructing Numerical Schemes**: Constructing numerical schemes for the neural differential equation generator driven by fBm and proving the convergence of the discretized solution. 3. **Criteria for Selecting Activation Functions**: Based on theoretical analysis, providing criteria for selecting activation functions for the driving neural network to make the continuous generation model achievable. 4. **Application Examples**: Constructing a generation model for generating artificial and real time series data. Experiments show that the calibrated model generator can well replicate the distribution characteristics of the original time series, especially the LRD characteristics. ### Summary By introducing fSDE-Net, the paper successfully addresses the problem of generating complex time series data with long-term memory characteristics. This method is not only rigorously proven theoretically but also shows good performance in practical applications, especially in data generation tasks in fields such as finance.

Fractional SDE-Net: Generation of Time Series Data with Long-term Memory

A deep learning-based Monte Carlo simulation scheme for stochastic differential equations driven by fractional Brownian motion

Statistical solution to SDEs with ˛ -stable Lévy noise via deep neural network

Neural Fractional Differential Equations

Statistical Solution to SDEs with $$\alpha $$-Stable Lévy Noise Via Deep Neural Network

Nonparametric Stochastic Differential Equations for Ultra-Short-Term Probabilistic Forecasting of Wind Power Generation

Stable Neural Stochastic Differential Equations in Analyzing Irregular Time Series Data

Time Series Forecasting with Ensembled Stochastic Differential Equations Driven by Lévy Noise

A Data-Driven Framework for Discovering Fractional Differential Equations in Complex Systems

Neural network stochastic differential equation models with applications to financial data forecasting

Finite Difference Neural Networks: Fast Prediction of Partial Differential Equations

FDE-Net: A Memory-Efficiency Densely Connected Network Inspired from Fractional-Order Differential Equations for Single Image Super-Resolution

SDHNet: a Sampling-Based Dual-Stream Hybrid Network for Long-Term Time Series Forecasting

Neural Rough Differential Equations for Long Time Series

SDE-HNN: Accurate and Well-calibrated Forecasting using Stochastic Differential Equations

SDE-Net: Equipping Deep Neural Networks with Uncertainty Estimates

Accurate and Reliable Forecasting using Stochastic Differential Equations

Neural SDEs as Infinite-Dimensional GANs

Neural Jump Stochastic Differential Equations

Deep Latent Regularity Network for Modeling Stochastic Partial Differential Equations

An Artificial Neural Network Approach for Solving Space Fractional Differential Equations