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.
Finite mixture models do not reliably learn the number of components
Authors: Diana Cai, Trevor Campbell, Tamara Broderick
ICML 2021 | Venue PDF | LLM Run Details | Input Tokens: 21,580 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 5,166 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 | We illustrate practical consequences of our theory on simulated and real data. |
| Researcher Affiliation | Academia | 1Department of Computer Science, Princeton University 2Department of Statistics, University of British Columbia 3CSAIL, Massachusetts Institute of Technology |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The provided code link (https://github.com/jwmi/Bayesian Mixtures.jl) is for a Julia implementation of a finite mixture model that the authors used for their experiments to illustrate their theory. It is not code that implements the novel theoretical methodology described in the paper. |
| Open Datasets | Yes | single-cell RNA sequencing data from mouse cortex and hippocampus cells (Zeisel et al., 2015) and m RNA expression data from human lung tissue (Bhattacharjee et al., 2001) |
| Dataset Splits | No | The paper describes generating datasets of increasing sizes and using subsets contained within larger ones, and discarding burn-in iterations, but does not specify explicit train/validation/test splits in terms of percentages or counts for data partitioning. |
| Hardware Specification | No | The paper mentions using a "Julia implementation" but does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions using a "Julia implementation of split-merge collapsed Gibbs sampling" and links to "Bayesian Mixtures.jl" but does not specify version numbers for Julia or the package. |
| Experiment Setup | Yes | For all experiments below, we use a finite mixture model with a multivariate Gaussian component family having diagonal covariance matrices and a conjugate prior on each dimension...We set the hyperparameters of the Bayesian finite mixture model as follows: m = 1 2(maxn [ N] Xn+minn [ N] Xn) where N = 10,000, κ = (maxn [ N] Xn minn [ N] Xn) 2, α = 2, r = 0.1, γ = 1, and β Gam(0.2, 10/κ). We ran a total of 100,000 Markov chain Monte Carlo iterations per data set; we discarded the first 10,000 iterations as burn-in. |