Research map / Time series / Causal and generative time series
K-ProtoDiff: Key Prototypes-Guided Diffusion for Time Series Generation
Yuhang Duan, Lin Lin, Xiaoshuai Wu
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 |
|---|---|---|---|
| use R-sampling instead of P-sampling (ablation) | K-ProtoDiff · Stocks | Predictive Score: 0.034 -> 0.025 | Improved |
| use KPL instead of MLP-based prototype module (ablation) | K-ProtoDiff with R-sampling · Stocks | Predictive Score: 0.138 -> 0.025 | Improved |
| add KPL prototype guidance (ablation) | K-ProtoDiff with R-sampling · EEG | Predictive Score: 0.584 -> 0.426 | Improved |
| add KPL to P-sampling variant (ablation) | K-ProtoDiff · Stocks | Predictive Score: 0.101 -> 0.034 | Improved |
| add R-sampling with prototype reflection on top of KPL (ablation) | K-ProtoDiff · EEG | Segment-wise DTW: 1.316 -> 1.207 | Improved |
| use K-ProtoDiff instead of Diffusion-TS | diffusion time series generation · Traffic (24 steps) | Context-FID: 0.818 -> 0.037 | Improved |
| use K-ProtoDiff instead of Diffusion-TS | diffusion time series generation · EEG (24 steps) | Discriminative Score: 0.053 -> 0.491 | Hurt |
| use K-ProtoDiff instead of Diffusion-TS | diffusion time series generation · Weather (segment length 6) | Segment-wise DTW: 2.259 -> 2.289 | No clear effect |
| vary number of prototypes and KAN order (ablation) | KPL · Energy | Predictive Score: best 0.128 at four prototypes, KAN order 2 | No clear effect |
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.
- Diffusion-TS — Diffusion-TS: Interpretable Diffusion for General Time Series Generation (ICLR 2024) · better on 7 of 9 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
- TimeDART: A Diffusion Autoregressive Transformer for Self-Supervised Time Series Representation (ICML 2025 · similarity 0.87)
- Time Series Generation Under Data Scarcity: A Unified Generative Modeling Approach (NeurIPS 2025 · similarity 0.87)
- ABC: Any-Subset Autoregression via Non-Markovian Diffusion Bridges in Continuous Time and Space (NeurIPS 2026 · similarity 0.87)
- Retrieval-Augmented Diffusion Models for Time Series Forecasting (NeurIPS 2024 · similarity 0.87)
- Dynamical Diffusion: Learning Temporal Dynamics with Diffusion Models (ICLR 2025 · similarity 0.87)
- TimeDP: Learning to Generate Multi-Domain Time Series with Domain Prompts (AAAI 2025 · similarity 0.87)
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.
