Research map / Time series / Causal and generative time series
Primal-Spectral Generative Modeling: Fast Analytical Generation via Pseudoinverse Lévy Inversion
Zhiyuan Ouyang, Simei Huang, Zuokai Wen, Xiangyun Zhang, 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 |
|---|---|---|---|
| use PriSpecNet-TS 50-NFE Heun instead of ARMD | time series generation · ETTh | Context-FID: 0.093 -> 0.018 | Improved |
| use PriSpecNet-TS 50-NFE Heun instead of ARMD | time series generation · Stock | Context-FID: 0.141 -> 0.052 | Improved |
| use PriSpecNet-TS 50-NFE Heun instead of ARMD | forecasting · Solar | MSE: 0.161 -> 0.113 | Improved |
| use PriSpecNet-TS 50-NFE Heun instead of ARMD | forecasting · Stock | MSE: 0.235 -> 0.179 | Improved |
| use 1-NFE PiLI instead of 50-NFE Heun sampler (ablation) | PriSpecNet-TS · ETTh | Context-FID: 0.018 -> 0.033 | Hurt |
| use 1-NFE PiLI instead of 25-NFE DPM-Solver++ (ablation) | PriSpecNet-XL · ImageNet 256x256 | FID: 1.62 -> 1.66 | Hurt |
| use 1-NFE PiLI instead of 25-NFE DPM-Solver++ (ablation) | PriSpecNet-XL · ImageNet 256x256 | Total Gflops: 4,436 -> 26 | Improved |
| use PiLI-XL instead of iMF-XL/2 | 1-NFE generation · ImageNet 256x256 | FID: 1.72 -> 1.66 | Improved |
| add spectral components with K=256 to PriSpecNet-B (early training) (ablation) | PriSpecNet-B · ImageNet 256x256 (200k steps) | FID: 128.53 -> 118.64 | Improved |
| increase spectral dimension K from 256 to 768 (ablation) | PriSpecNet-B · ImageNet 256x256 (final) | FID: 98.09 -> 48.00 | Improved |
| increase spectral dimension K from 768 to 1280 (ablation) | PriSpecNet-B · ImageNet 256x256 | FID: 48.00 -> above 140 | Hurt |
| increase spectral dimension K from 768 to 1280 (ablation) | PriSpecNet-XL · ImageNet 256x256 (final) | FID: 14.53 -> 1.62 | Improved |
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 4 of 4 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
- Free Decompression with Algebraic Spectral Curves (NeurIPS 2026 · similarity 0.85)
- Manifolds, Random Matrices and Spectral Gaps: The geometric phases of generative diffusion (ICLR 2025 · similarity 0.85)
- The Geometry of Noise: Why Diffusion Models Don't Need Noise Conditioning (NeurIPS 2026 · similarity 0.85)
- Unified Framework for Diffusion Generative Models in SO(3): Applications in Computer Vision and Astrophysics (AAAI 2024 · similarity 0.85)
- Generative Modeling on Manifolds Through Mixture of Riemannian Diffusion Processes (ICML 2024 · similarity 0.85)
- Rectifying Categorical Flows on Statistical Manifolds for One-Step Generation (NeurIPS 2026 · 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.
