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..
A Black-Box Debiasing Framework for Conditional Sampling
Authors: Han Cui, Jingbo Liu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we provide numerical experiments to illustrate the debiasing framework for posterior approximation under the binary prior case and the Gaussian mixture prior case. For both settings, we examine the convergence rate of the debiased estimators Dn,kg(T/n) for k = 1, 2, 3, 4. The results are shown in log-log plots in Figure 1, where the vertical axis represents the logarithm of the absolute error and the horizontal axis represents the logarithm of the sample size n. The results are shown in Figure 2. The figure presents log-log plots where the vertical axis represents the logarithm of the absolute error or of the variance and the horizontal axis represents the logarithm of the sample size n. |
| Researcher Affiliation | Academia | Han Cui University of Illinois at Urbana-Champaign Champaign, IL EMAIL Jingbo Liu University of Illinois at Urbana-Champaign Champaign, IL EMAIL |
| Pseudocode | Yes | Algorithm 1 Posterior Approximation via Debiasing Framework Algorithm 2 Rejection Sampling for Debiased Posterior e PXn(x | y = y ) |
| Open Source Code | Yes | Neur IPS Paper Checklist 5. Open access to data and code 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: We used the simulation data in the experiment. |
| Open Datasets | No | The paper uses simulated data defined within the text: "Binary prior case. Suppose that X = {0, 1} and X Bern(q) for some unknown prior q (0, 1)..." and "Gaussian mixture prior case. Suppose that X 1 2N(0, 1) + 1 2N(1, 1) and Y = X + ξ where ξ N(0, 1/16)." No external publicly available datasets are utilized with concrete access information. |
| Dataset Splits | No | The paper uses simulated data and describes the generation of 'n i.i.d. samples D = {Xi}n i=1'. It does not specify predefined training, test, or validation splits for a fixed dataset, as it's not evaluating against a pre-existing dataset. |
| Hardware Specification | No | Neur IPS Paper Checklist 8. Experiments compute resources Question: For each experiment, does the paper provide sufficient information on the computer resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [No] Justification: Our experiment is a simple simulation for binary case. |
| Software Dependencies | No | The paper does not explicitly mention any software dependencies with specific version numbers (e.g., Python 3.8, PyTorch 1.9, etc.). |
| Experiment Setup | Yes | Binary prior case. Suppose that X = {0, 1} and X Bern(q) for some unknown prior q (0, 1). ...in the first experiment we set q = 0.4, y = 2, and Y |X N(X, 1), while in the second we set q = 3/11, y = 1, and Y |X N(X, 1/4). Gaussian mixture prior case. Suppose that X 1 2N(0, 1) + 1 2N(1, 1) and Y = X + ξ where ξ N(0, 1/16). Additionally, let y = 0.8 and A = {x : x 0.5}. To ensure that the Monte Carlo error is negligible compared to the bias O(n k), we select the number of Monte Carlo samples N such that N n2k 1. In practice, we run simulations for k = 1 and k = 2 and set N = n3 for k = 1 and N = n4 for k = 2. |