RECORD
On the Constrained Time-Series Generation Problem
Machine Learning · Submitted 4 Jul 2023
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim Most importantly, we show that our "GuidedDiffTime" model is the only solution where re-training is not necessary for new constraints, resulting in a significant carbon footprint reduction, up to 92% w.r.t. existing deep learning methods.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Synthetic time series are often used in practical applications to augment the historical time series dataset for better performance of machine learning algorithms, amplify the occurrence of rare events, and also create counterfactual scenarios described by the time series. Distributional-similarity (which we refer to as realism) as well as the satisfaction of certain numerical constraints are common requirements in counterfactual time series scenario generation requests. For instance, the US Federal Reserve publishes synthetic market stress scenarios given by the constrained time series for financial institutions to assess their performance in hypothetical recessions. Existing approaches for generating constrained time series usually penalize training loss to enforce constraints, and reject non-conforming samples. However, these approaches would require re-training if we change constraints, and rejection sampling can be computationally expensive, or impractical for complex constraints. In this paper, we propose a novel set of methods to tackle the constrained time series generation problem and provide efficient sampling while ensuring the realism of generated time series. In particular, we frame the problem using a constrained optimization framework and then we propose a set of generative methods including "GuidedDiffTime", a guided diffusion model to generate realistic time series. Empirically, we evaluate our work on several datasets for financial and energy data, where incorporating constraints is critical. We show that our approaches outperform existing work both qualitatively and quantitatively. Most importantly, we show that our "GuidedDiffTime" model is the only solution where re-training is not necessary for new constraints, resulting in a significant carbon footprint reduction, up to 92% w.r.t. existing deep learning methods.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Synthetic time series are often used in practical applications to augment the historical time series dataset for better performance of machine learning algorithms, amplify the occurrence of rare events, and also create counterfactual scenarios described by the time series. | background 1.00 | 0.04 | 0.02 | asserted 0.97 |
| s1 | Distributional-similarity (which we refer to as realism) as well as the satisfaction of certain numerical constraints are common requirements in counterfactual time series scenario generation requests. | background 1.00 | 0.06 | 0.02 | asserted 0.99 |
| s2 | For instance, the US Federal Reserve publishes synthetic market stress scenarios given by the constrained time series for financial institutions to assess their performance in hypothetical recessions. | background 1.00 | 0.05 | 0.02 | asserted 1.00 |
| s3 | Existing approaches for generating constrained time series usually penalize training loss to enforce constraints, and reject non-conforming samples. | background 0.99 | 0.03 | 0.02 | asserted 0.99 |
| s4 | However, these approaches would require re-training if we change constraints, and rejection sampling can be computationally expensive, or impractical for complex constraints. | gap 0.85 | 0.08 | 0.03 | asserted 0.57 |
| s5 | In this paper, we propose a novel set of methods to tackle the constrained time series generation problem and provide efficient sampling while ensuring the realism of generated time series. | contribution 1.00 | 0.83 | 0.15 | asserted 0.98 |
| s6 | In particular, we frame the problem using a constrained optimization framework and then we propose a set of generative methods including "GuidedDiffTime", a guided diffusion model to generate realistic time series. | method 0.85 | 0.92 | 0.07 | asserted 0.99 |
| s7 | Empirically, we evaluate our work on several datasets for financial and energy data, where incorporating constraints is critical. | method 0.84 | 0.90 | 0.03 | asserted 0.86 |
| s8 | We show that our approaches outperform existing work both qualitatively and quantitatively. | result 1.00 | 0.96 | 0.18 | asserted 1.00 |
| s9 | Most importantly, we show that our "GuidedDiffTime" model is the only solution where re-training is not necessary for new constraints, resulting in a significant carbon footprint reduction, up to 92% w.r.t. existing deep learning methods. | result 0.97 | 0.95 | 0.78 | asserted 1.00 |
What was examined, and what was not
- arXiv listing abstract v2
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Synthetic time series are often used in practical applications to augment the historical time series dataset for better performance of machine learning algorithms, amplify the occurrence of rare events, and also create counterfactual scenarios described by the time series. Distributional-similarity (which we refer to as realism) as well as the satisfaction of certain numerical constraints are common requirements in counterfactual time series scenario generation requests. For instance, the US Federal Reserve publishes synthetic market stress scenarios given by the constrained time series for financial institutions to assess their performance in hypothetical recessions. Existing approaches for generating constrained time series usually penalize training loss to enforce constraints, and reject non-conforming samples. However, these approaches would require re-training if we change constraints, and rejection sampling can be computationally expensive, or impractical for complex constraints. In this paper, we propose a novel set of methods to tackle the constrained time series generation problem and provide efficient sampling while ensuring the realism of generated time series. In particular, we frame the problem using a constrained optimization framework and then we propose a set of generative methods including "GuidedDiffTime", a guided diffusion model to generate realistic time series. Empirically, we evaluate our work on several datasets for financial and energy data, where incorporating constraints is critical. We show that our approaches outperform existing work both qualitatively and quantitatively. Most importantly, we show that our "GuidedDiffTime" model is the only solution where re-training is not necessary for new constraints, resulting in a significant carbon footprint reduction, up to 92% w.r.t. existing deep learning methods.
Human review
How the literature uses it
6 arXiv papers cite this in 7 sentences
Papers citing it · 6 sentences without a confident label
Compares with it 1 paper
-
For time series generation, we select four previous models to compare with: TimeVAE [ref] , Diffwave [ref] , TimeGAN [ref] , Cot-GAN [ref] and DiffTime [ref] which can be viewed as an unconditional CSDI [ref] .
Diffusion-TS: Interpretable Diffusion for General Time Series Generation · Empirical Evaluation · 1.00
What it cites
Compares with 5 papers
-
Baselines We compare our approaches against existing TS generative models, including GT-GAN [ref] , TimeGAN [ref] , RCGAN [ref] , C-RNN-GAN [ref] , a Recurrent Neural Networks (RNN) [ref] trained with T-Forcing and P-Forcing [ref] , WaveNET [ref] , and WaveGAN [ref] .
Generating Sequences With Recurrent Neural Networks · Experiments · 0.98
-
Baselines We compare our approaches against existing TS generative models, including GT-GAN [ref] , TimeGAN [ref] , RCGAN [ref] , C-RNN-GAN [ref] , a Recurrent Neural Networks (RNN) [ref] trained with T-Forcing and P-Forcing [ref] , WaveNET [ref] , and WaveGAN [ref] .
WaveNet: A Generative Model for Raw Audio · Experiments · 0.98
-
Baselines We compare our approaches against existing TS generative models, including GT-GAN [ref] , TimeGAN [ref] , RCGAN [ref] , C-RNN-GAN [ref] , a Recurrent Neural Networks (RNN) [ref] trained with T-Forcing and P-Forcing [ref] , WaveNET [ref] , and WaveGAN [ref] .
C-RNN-GAN: Continuous recurrent neural networks with adversarial training · Experiments · 0.98
Cites as background 2 papers
-
Related work: Existing work employs deep generative models (DGMs) to capture the statistical data properties and temporal dynamics of TS [ref] , and additional constraints are usually introduced by penalizing the generative model proportionally to the mass it allocates to invalid data [ref] ; or by adding a regularization term to the loss function [ref] .
Objective-Reinforced Generative Adversarial Networks (ORGAN) for Sequence Generation Models · Introduction · 0.93
-
Finally, a different line of work proposes special-purpose solutions, with ad-hoc architectures or sampling methods, which however tackle specific applications (not TS generation) [ref] .
MolGAN: An implicit generative model for small molecular graphs · Introduction · 0.87
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.