Research map / Time series / Causal and generative time series
MMPD: Diverse Time Series Forecasting via Multi-Mode Patch Diffusion Loss
Yunhao Zhang, Wenyao Hu, Jiale Zheng, Lujia Pan, Junchi Yan
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 adjacent-patch conditioning (r>=1) to the denoising MLP (ablation) | MMPD Patch Consistent MLP · Dynamic tau=180 | Top-3 MSE / Top-3 MAE | Improved |
| increase adjacent range r beyond 1 (ablation) | Patch Consistent MLP · Dynamic tau=180 | Top-3 MSE | Improved |
| add deterministic anchor term (lower lambda from 1.0 to 0.999) (ablation) | MMPD loss · Dynamic tau=180 | MSE | Improved |
| lower lambda from 1.0 to 0.9 (ablation) | MMPD loss · Dynamic tau=180 | Top-3 MSE, CRPS | Improved |
| use evolving variational GMM instead of post-hoc GMM (ablation) | multi-mode inference · Dynamic tau=180 | Top-3 MSE: 0.310 -> 0.301 | Improved |
| use evolving variational GMM instead of KMeans (ablation) | multi-mode inference · Dynamic tau=180 | Top-3 MSE: 0.324 -> 0.301 | Improved |
| use evolving variational GMM instead of random assignment (ablation) | multi-mode inference · Dynamic tau=180 | Top-3 MSE: 0.365 -> 0.301 | Improved |
| replace Mix loss with MMPD loss | decoder-only Transformer · 8 datasets | Top-3 MSE avg rank: 1.875 -> 1 | Improved |
| replace MSE loss with MMPD loss | decoder-only Transformer · 8 datasets | MSE avg rank: 1.875 vs 1.75 | No clear effect |
| replace Student-T loss with MMPD loss | decoder-only Transformer · 8 datasets | CRPS avg rank: 2.25 vs 2 | No clear effect |
| apply MMPD instead of MSE or Mix | Crossformer, SegRNN, MaskAE · 8 datasets | Top-3 MSE / Top-3 MAE | Improved |
| use MMPD instead of Mix (CRPS stability) | SegRNN · 8 datasets | CRPS: Mix Inf on all datasets | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CSDI — CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation (NeurIPS 2021) · better on 4 of 4 shared results
- Diffusion-TS — Diffusion-TS: Interpretable Diffusion for General Time Series Generation (ICLR 2024) · better on 1 of 1 shared result
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
- Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting (AAAI 2025 · similarity 0.88)
- Unlocking the Power of Patch: Patch-Based MLP for Long-Term Time Series Forecasting (AAAI 2025 · similarity 0.87)
- MG-TSD: Multi-Granularity Time Series Diffusion Models with Guided Learning Process (ICLR 2024 · similarity 0.87)
- DiffPTS: Rethinking Diffusion ELBO for Probabilistic Time Series Forecasting (NeurIPS 2026 · similarity 0.87)
- Dirichlet-Guided Group Forecasting for Alleviating Over-smoothing in Time Series Forecasting (NeurIPS 2026 · similarity 0.86)
- Non-stationary Diffusion For Probabilistic Time Series Forecasting (ICML 2025 · similarity 0.86)
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.
