Research map / Time series / Causal and generative time series
Diff-MN: Diffusion Parameterized MoE-NCDE for Continuous Time Series Generation with Irregular Observations
Xu Zhang, Junwei Deng, Chang Xu, Hao Li, Jiang Bian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add mixture-of-experts dynamics function (ablation) | Diff-MN · ECG5K (30% dropped) | MDD: 0.152 -> 0.062 | Improved |
| add decoupled design with pretrained channel-wise autoencoder (ablation) | Diff-MN · ECG5K (30% dropped) | MDD: 0.191 -> 0.062 | Improved |
| combine MoE and decoupled design (ablation) | Diff-MN · ECGFD (30% dropped) | MDD: 0.965 -> 0.424 | Improved |
| use four experts rather than one expert with same layers (ablation) | MoE-NCDE · ECG200 (30% dropped) | KL divergence: 0.136 -> 0.052 | Improved |
| use four experts rather than one expert with same layers (ablation) | MoE-NCDE · ECG5K (30% dropped) | Discriminative score: 0.364 -> 0.205 | Improved |
| use diffusion-generated MoE weights for continuous generation | Diff-MN · Sines continuous forecasting (30% dropped) | MSE: 0.049 -> 0.026 | Improved |
| apply standard NCDE for continuous generation | KoVAE · Stocks continuous forecasting (30% dropped) | MSE: 0.025 -> 0.041 | Hurt |
| use dense MoE instead of sparse MoE (ablation) | MoE-NCDE · ECG datasets | Discriminative score: 7 vs 5 wins | Mixed |
| increase number of experts beyond one (ablation) | MoE-NCDE · ECG datasets | MDD: one expert 0.176 -> four experts 0.079 | Mixed |
| use MoE dynamics over standard NCDE in time cost | MoE-NCDE · Stocks length 12 | generation time: 0.81s -> 1.2s | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GT-GAN — GT-GAN: General Purpose Time Series Synthesis with Generative Adversarial Networks (NeurIPS 2022) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Non-isotropic Time Series Diffusion Model with Moving Average Transitions (ICML 2025 · similarity 0.87)
- KoNODE: Koopman-Driven Neural Ordinary Differential Equations with Evolving Parameters for Time Series Analysis (ICML 2025 · similarity 0.86)
- Conditional Diffusion Model with Nonlinear Data Transformation for Time Series Forecasting (ICML 2025 · similarity 0.85)
- Diffusion-TS: Interpretable Diffusion for General Time Series Generation (ICLR 2024 · similarity 0.85)
- Auto-Regressive Moving Diffusion Models for Time Series Forecasting (AAAI 2025 · similarity 0.85)
- On the Generalization and Approximation Capacities of Neural Controlled Differential Equations (ICLR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
