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..
Constrained Posterior Sampling: Time Series Generation with Hard Constraints
Authors: Sai Shankar Narasimhan, Shubhankar Agarwal, Litu Rout, Sanjay Shakkottai, Sandeep Chinchali
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, CPS outperforms state-of-the-art methods in sample quality and similarity to real time series by around 70% and 22%, respectively, on real-world stocks, traffic, and air quality datasets. ... 5 Experiments In this section, we describe the experiments, datasets, baselines, and metrics used to evaluate CPS. |
| Researcher Affiliation | Academia | Chandra Family Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, TX, 78712. EMAIL |
| Pseudocode | Yes | Algorithm 1 Constrained Posterior Sampling |
| Open Source Code | Yes | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: The supplementary material contains the data and code used for experiments. |
| Open Datasets | Yes | Datasets: We use real-world datasets from the Stocks [7], Air Quality [41], and Traffic [42] domains. ... [41] S. Chen. Beijing Multi-Site Air-Quality Data. UCI Machine Learning Repository. DOI: https://doi.org/10.24432/C5RK5G. 2019. [42] J. Hogue. Metro Interstate Traffic Volume. UCI Machine Learning Repository. DOI: https://doi.org/10.24432/C5X60B. 2019. |
| Dataset Splits | Yes | The Air Quality dataset is a multivariate dataset with six channels. The total numbers of train, val, and test samples are 12166, 1537, and 1525, respectively. The Traffic dataset is univariate. The total train, val, and test samples are 1604, 200, and 201, respectively. The Stocks dataset is a multivariate dataset with six channels. The total train, val, and test samples are 2871, 358, and 360, respectively. The truncated form of the waveforms dataset used for evaluation consists of 13320, 1665, and 1665 train, val, and test samples, respectively. |
| Hardware Specification | Yes | Table 9 reports inference latency on a single NVIDIA RTX 6000 GPU. ... For each dataset, we trained the diffusion model on a single NVIDIA RTX 6000 GPU. |
| Software Dependencies | No | For the CPS implementation, we use CVXPY [35]. ... As suggested in [12], we use the SLSQP solver from Sci Py [58]. |
| Experiment Setup | Yes | We use 256 channels in each residual layer, with 16-dimensional vectors representing each channel. The diffusion time step input embedding is a 256-dimensional vector. Further, the metadata encoder has an embedding size of 256 for the conditional case. The metadata encoder has two attention layers with eight attention heads. All our experiments use a learning rate of 10^-4. Our training procedure and the hyperparameters are precisely the same as those in [1]. ... we trained the TIME WEAVER-CSDI denoiser up to a maximum of 5000 epochs for all datasets. |