Time Series Synthesis via Multi-scale Patch-based Generation of Wavelet Scalogram

TL;DR

This paper introduces a novel framework for generating synthetic time series by learning from a single example using wavelet scalogram patches, effective especially for trendless data.

Contribution

It presents a new patch-based generative approach for time series synthesis from minimal data, leveraging wavelet scalogram analysis.

Findings

Effective for trendless time series

Performs well with reshuffled samples of same duration

Less effective for retargeted longer sequences

Abstract

A framework is proposed for the unconditional generation of synthetic time series based on learning from a single sample in low-data regime case. The framework aims at capturing the distribution of patches in wavelet scalogram of time series using single image generative models and producing realistic wavelet coefficients for the generation of synthetic time series. It is demonstrated that the framework is effective with respect to fidelity and diversity for time series with insignificant to no trends. Also, the performance is more promising for generating samples with the same duration (reshuffling) rather than longer ones (retargeting).