Notice: This website is continuously updated as new conference proceedings and journal volumes are published. Consequently, the figures presented here may differ from those reported in our publications.
A brief overview of our methodology is provided on the Methods page, along with references to publications documenting our approach and links to the code and data used in each study.
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.
Stein $\Pi$-Importance Sampling
Authors: Congye Wang, Ye Chen, Heishiro Kanagawa, Chris J. Oates
NeurIPS 2023 | Venue PDF | LLM Run Details | Input Tokens: 38,495 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 4,194 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The approach is stress-tested using the recently released Posterior DB suite of benchmark tasks in Section 4, before concluding with a discussion in Section 5. |
| Researcher Affiliation | Academia | Congye Wang1, Wilson Ye Chen2, Heishiro Kanagawa1, Chris. J. Oates1 1 Newcastle University, UK 2 University of Sydney, Australia |
| Pseudocode | Yes | Algorithm 1 Π-Invariant Metropolis-Adjusted Langevin Algorithm (MALA); Algorithm 2 Stein Π-Importance Sampling (SΠIS-MALA); Algorithm 3 Stein Π-Thinning (SΠT-MALA) |
| Open Source Code | Yes | All experiments that we report can be reproduced using code available at https://github.com/congyewang/Stein-Pi-Importance-Sampling. |
| Open Datasets | Yes | To introduce objectivity into our assessment, we exploited the recently released Posterior DB benchmark (Magnusson et al., 2022). |
| Dataset Splits | No | The paper does not describe dataset splits for training, validation, or testing. |
| Hardware Specification | No | The paper does not explicitly describe the hardware used to run the experiments. |
| Software Dependencies | Yes | The linearly-constrained quadratic programme in Algorithm 2 was solved using the Python v3.10.4 packages qpsolvers v3.4.0 and proxsuite v0.3.7. [...] The test problems in Posterior DB are defined in the Stan probabilistic programming language, and so Bridge Stan (Roualdes et al., 2023) was used to directly access posterior densities and their gradients as required. [...] The version of Stan that we used was Stanc3 Version 2.31.0 (Unix). |
| Experiment Setup | Yes | For all experiments that we report using MALA, we set ϵ0 = 1, M0 = Id, h = 10, and α1 = = α9 = 0.3. The warm-up epoch lengths were n0 = = n8 = 1, 000 and the final epoch length was n9 = 105. |