Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Single-Step Operator Learning for Conditioned Time-Series Diffusion Models
Authors: Hui Chen, Vikas Singh
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Evaluations on multiple datasets show that our single-step generation proposal achieves forecasting/imputation results comparable (or superior) to many multi-step diffusion schemes while significantly reducing inference costs. Our code is available at: https://github.com/vsingh-group/SSOL-timeseries. |
| Researcher Affiliation | Academia | Hui Chen Department of Statistics University of Wisconsin Madison EMAIL Vikas Singh Department of Biostatistics and Medical Informatics University of Wisconsin Madison EMAIL |
| Pseudocode | Yes | Alg. 1 and Alg. 2 in Appendix B show the complete training and sampling procedures. |
| Open Source Code | Yes | Our code is available at: https://github.com/vsingh-group/SSOL-timeseries. |
| Open Datasets | Yes | All datasets used are publicly available and cited in Appendix C. The implementation code is available at https://github.com/vsingh-group/SSOL-timeseries. |
| Dataset Splits | Yes | Our experimental configurations follow the protocols established in Wu et al. [2023], including identical data processing and splitting methods. Details for each dataset are provided in Table 4. Aligned with fair comparison settings outlined in Liu et al. [2024], Li et al. [2025], we fix the lookback window length to 96 time steps for most datasets and baselines, with the prediction horizon set to 192 time steps. For imputation tasks, we fix the window length to 48 time steps, following the protocol in Yuan and Qiao [2024]. |
| Hardware Specification | Yes | All experiments were conducted using Py Torch Paszke et al. [2019] on a single NVIDIA A100 40GB GPU. |
| Software Dependencies | No | The paper mentions 'Py Torch' but does not specify a version number. While it cites Paszke et al. [2019], this refers to the publication of PyTorch rather than the specific version used in the experiments. |
| Experiment Setup | Yes | All experiments were conducted using Py Torch Paszke et al. [2019] on a single NVIDIA A100 40GB GPU. We trained our model using a two-stage approach with the Adam optimizer Kingma and Ba [2015]. In the first stage, we trained the model using the EDM-type loss Karras et al. [2022] defined in Eq. 4. The second stage focused on optimizing the composition property loss, as shown in Algorithm 1. Summary of the experimental configurations for all datasets is listed in Table 5. |