Unconditional flow-based time series generation with equivariance-regularised latent spaces

TL;DR

This paper introduces a novel flow-based time series generation method that uses equivariance-regularised latent spaces, leading to improved quality and faster sampling compared to existing diffusion-based models.

Contribution

It proposes a simple regularisation technique for autoencoders to enforce equivariance in latent spaces, enhancing time-series generation quality and efficiency.

Findings

Outperforms diffusion-based baselines in quality metrics

Achieves orders-of-magnitude faster sampling

Demonstrates benefits of geometric inductive biases in latent models

Abstract

Flow-based models have proven successful for time-series generation, particularly when defined in lower-dimensional latent spaces that enable efficient sampling. However, how to design latent representations with desirable equivariance properties for time-series generative modelling remains underexplored. In this work, we propose a latent flow-matching framework in which equivariance is explicitly encouraged through a simple regularisation of a pre-trained autoencoder. Specifically, we introduce an equivariance loss that enforces consistency between transformed signals and their reconstructions, and use it to fine-tune latent spaces with respect to basic time-series transformations such as translation and amplitude scaling. We show that these equivariance-regularised latent spaces improve generation quality while preserving the computational advantages of latent flow models. Experiments on multiple real-world datasets demonstrate that our approach consistently outperforms existing diffusion-based baselines in standard time-series generation metrics, while achieving orders-of-magnitude faster sampling. These results highlight the practical benefits of incorporating geometric inductive biases into latent generative models for time series.