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..
Smooth Sailing: Lipschitz-Driven Uncertainty Quantification for Spatial Associations
Authors: David Burt, Renato Berlinghieri, Stephen Bates, Tamara Broderick
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In simulated and real data experiments, we find that our method consistently achieves nominal coverage, whereas all the alternatives dramatically fail to do so. We also provide ablation studies to evaluate the effect of varying the Lipschitz constant in both simulated and real settings. |
| Researcher Affiliation | Academia | David R. Burt Renato Berlinghieri Stephen Bates Tamara Broderick MIT LIDS EMAIL |
| Pseudocode | Yes | We detail how to efficiently compute our proposed confidence interval for θ OLS,p in Algorithm 1 (Appendix B). We prove its validity in Theorem 7 below. |
| Open Source Code | Yes | Anonymized code is available at https://anonymous.4open.science/r/LDI-Neur IPS-4EEB. |
| Open Datasets | Yes | We use the 983 data points from Lu et al. [37], who in turn draw on [68, 65, 43]. We define our target region in the West portion of CONUS as locations with latitude in the range (25, 50) and longitude in the range (-125, -110). ... We use the 983 data points from Lu et al. [37] ... USFS Tree Canopy Cover (TCC) product [68]. This dataset is public domain. ... Global Aridity Index (1970 2000): Averaged at a 30 arc-seconds resolution [65]. ... Elevation: Provided by NASA s 30-meter resolution dataset [43]. ... Slope: NASA s 30 m Digital Elevation Model. Also provided by NASA s 30-meter resolution dataset [43]. While we could not find specific license information, as the Slope and Elevation datasets are produced by a US government agency (NASA), we understand this data to be public domain following section 105 of the Copyright Act of 1976. |
| Dataset Splits | Yes | Out of all points in this region, we designate 50% totaling 133 sites as target data. Next, we select the source data by taking a uniform random sample of 20% of the remaining spatial locations, repeated over 250 random seeds to assess coverage performance. Each seed yields 170 source locations. |
| Hardware Specification | Yes | All simulation experiments were run on a Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz using 36 threads. The total time to run all simulation experiments was under two hours. ... The tree cover experiments was run on a Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz using 36 threads. The total time to run the experiment was under 5 minutes. |
| Software Dependencies | Yes | In practice, we use the Scipy sparse matrix algebra [72] and the CLARABEL solver [22] through the CVXPY optimization interface [16, 1] to solve this quadratic program. ... Sci Py 1.0: Fundamental Algorithms for Scientific Computing in Python. |
| Experiment Setup | Yes | We vary shift [0, 0.2, 0.4, 0.6, 0.8] and run 250 seeds for each shift. ... Now we repeat the first simulation but vary L {0.1, 0.5, 1.0, 2.0, 3.5, 5, 7.5, 10}. ... For the tree cover experiment, we leverage domain knowledge to set the Lipschitz constant to L = 0.2, in units of percent tree cover per kilometer (km). ... For the simulation experiments, we performed cross-validation to select the bandwidth parameters over the set {0.01, 0.025, 0.05, 0.1, 0.25, 0.5}. ... We maximize the likelihood of Y N(θTX, Σ), with Σ specified by an isotropic Matérn 3/2 covariance function and a nugget, to select the parameters of the covariance function and nugget variance. |