Abstract:Denoising diffusion probabilistic models (DDPMs) are becoming the leading paradigm for generative models. It has recently shown breakthroughs in audio synthesis, time series imputation and forecasting. In this paper, we propose Diffusion-TS, a novel diffusion-based framework that generates multivariate time series samples of high quality by using an encoder-decoder transformer with disentangled temporal representations, in which the decomposition technique guides Diffusion-TS to capture the semantic meaning of time series while transformers mine detailed sequential information from the noisy model input. Different from existing diffusion-based approaches, we train the model to directly reconstruct the sample instead of the noise in each diffusion step, combining a Fourier-based loss term. Diffusion-TS is expected to generate time series satisfying both interpretablity and realness. In addition, it is shown that the proposed Diffusion-TS can be easily extended to conditional generation tasks, such as forecasting and imputation, without any model changes. This also motivates us to further explore the performance of Diffusion-TS under irregular settings. Finally, through qualitative and quantitative experiments, results show that Diffusion-TS achieves the state-of-the-art results on various realistic analyses of time series.

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are several key challenges in high - quality time - series generation, specifically including: 1. **High - quality time - series generation**: Existing diffusion models often lack the ability to capture the complex dynamic characteristics of time - series when generating time - series, especially when dealing with multivariate and long - term time - series. These problems lead to the low quality of the generated time - series, which is difficult to meet the needs of practical applications. 2. **Interpretability and authenticity**: Existing methods often have difficulty ensuring the interpretability and authenticity of the generated results simultaneously when generating time - series. That is, the generated time - series not only needs to look real, but also needs to be able to explain each component in its generation process, such as trends, seasonality, and errors. 3. **Flexibility of conditional generation tasks**: Existing time - series generation models usually need specific modifications or adjustments to the model when dealing with conditional generation tasks (such as prediction and imputation), which limits the flexibility and generality of the model. To address these challenges, the paper proposes a new diffusion framework named **Diffusion - TS**. This framework aims to generate high - quality multivariate time - series samples by introducing decoupled time representations (such as trend and seasonality decomposition) and a Transformer - based encoder - decoder structure. In addition, Diffusion - TS also designs a Fourier - transform - based loss function to directly reconstruct samples instead of noise, thereby improving the quality of the generated time - series. The paper also demonstrates the flexibility of Diffusion - TS in conditional generation tasks, which can be applied to tasks such as prediction and imputation without any changes to the model. In summary, the main contributions of the paper are: - Proposing a time - series generation framework, Diffusion - TS, which combines season - trend decomposition techniques and denoising diffusion models. - Designing a Fourier - transform - based training objective, enabling the model to generate time - series more accurately. - Achieving flexible conditional generation tasks through instance - aware guidance strategies. - Experimental results show that Diffusion - TS is competitive in generating high - quality time - series and performs well in terms of interpretability and flexibility.

Diffusion-TS: Interpretable Diffusion for General Time Series Generation

TS-Diffusion: Generating Highly Complex Time Series with Diffusion Models

Denoising Diffusion Straightforward Models for Energy Conversion Monitoring Data Imputation

Probabilistic Time Series Modeling with Decomposable Denoising Diffusion Model

Non-autoregressive Conditional Diffusion Models for Time Series Prediction

TimeDiT: General-purpose Diffusion Transformers for Time Series Foundation Model

Diffusion Models for Time Series Applications: A Survey

Predict, Refine, Synthesize: Self-Guiding Diffusion Models for Probabilistic Time Series Forecasting

SeqDiffuSeq: Text Diffusion with Encoder-Decoder Transformers

Generative Time Series Forecasting with Diffusion, Denoise, and Disentanglement

Think While You Generate: Discrete Diffusion with Planned Denoising

Latent Diffusion Transformer for Probabilistic Time Series Forecasting

Population Aware Diffusion for Time Series Generation

MG-TSD: Multi-granularity time series diffusion models with guided learning process

UTSD: Unified Time Series Diffusion Model

Improving Efficiency of Diffusion Models via Multi-Stage Framework and Tailored Multi-Decoder Architectures

Diffusion Models for Tabular Data Imputation and Synthetic Data Generation

Stochastic Diffusion: A Diffusion Probabilistic Model for Stochastic Time Series Forecasting

Generation of synthetic financial time series by diffusion models

Diffusion with Forward Models: Solving Stochastic Inverse Problems Without Direct Supervision