Research map / Time series / Causal and generative time series
Latent-to-Data Cascaded Diffusion Models for Unconditional Time Series Generation
Lifeng Shen, Kai Syun Hou, Weiyu Chen, James Kwok
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 data-space diffusion after latent diffusion (ablation) | L2D-Diff latent-only · Stock | C-FID: 3.682 -> 0.310 | Improved |
| add data-space diffusion after latent diffusion (ablation) | L2D-Diff latent-only · Character Trajectories | C-FID: 1.829 -> 0.284 | Improved |
| add latent-space conditioning to data-space diffusion (ablation) | L2D-Diff data-only · Stock | C-FID: 0.385 -> 0.310 | Improved |
| add latent-space conditioning to data-space diffusion (ablation) | L2D-Diff data-only · Character Trajectories | C-FID: 2.368 -> 0.284 | Improved |
| use CNN instead of MLP for conditioning network F (ablation) | L2D-Diff · Character Trajectories | C-FID: 0.43 -> 0.28 | Improved |
| use depth 5 instead of 1 layer in conditioning network (ablation) | L2D-Diff · Character Trajectories | C-FID: 0.58 -> 0.28 | Improved |
| increase conditioning network depth from 5 to 20 layers (ablation) | L2D-Diff · Stocks | C-FID: 0.31 -> 0.75 | Hurt |
| use concatenation instead of cross-attention for latent conditioning (ablation) | L2D-Diff · Japanese Vowels / Character Trajectories | C-FID: 0.61 -> 0.28 | Improved |
| use concatenation instead of direct addition for latent conditioning (ablation) | L2D-Diff · Stocks | C-FID: 0.55 -> 0.31 | Improved |
| use concatenation instead of FiLM (ablation) | L2D-Diff · Stocks | C-FID: 0.32 -> 0.31 | No clear effect |
| increase latent dimension from 8 to 64 (ablation) | L2D-Diff · Stock | C-FID: 0.310 -> 1.121 | Hurt |
| increase latent dimension from 8 to 128 (ablation) | L2D-Diff · Character Trajectories | C-FID: 0.284 -> 2.947 | 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.
- Diffusion-TS — Diffusion-TS: Interpretable Diffusion for General Time Series Generation (ICLR 2024) · better on 11 of 11 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
- Diffusion-TS: Interpretable Diffusion for General Time Series Generation (ICLR 2024 · similarity 0.89)
- Time Series Generation Under Data Scarcity: A Unified Generative Modeling Approach (NeurIPS 2025 · similarity 0.89)
- Population Aware Diffusion for Time Series Generation (AAAI 2025 · similarity 0.89)
- A Diffusion Model for Regular Time Series Generation from Irregular Data with Completion and Masking (NeurIPS 2025 · similarity 0.88)
- SigDiffusions: Score-Based Diffusion Models for Time Series via Log-Signature Embeddings (ICLR 2025 · similarity 0.88)
- TimeDART: A Diffusion Autoregressive Transformer for Self-Supervised Time Series Representation (ICML 2025 · similarity 0.88)
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.
