Learning Parametric Distributions from Samples and Preferences
Authors: Marc Jourdan, Gizem Yüce, Nicolas Flammarion
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 6. Experiments In this section, we compare the empirical performance of the different estimators introduced in this paper. For preferences based on rθ = log pθ, we conduct a set of experiments for Gaussian distributions, and defer to Appendix H.1 for experiments on Laplace and Rayleigh distributions. In particular, we consider a uniformly drawn mean parameter θ U([1, 2]d) and the isotropic covariance Σ = Id. For sample size n [Nmax] with Nmax = 104, we compute the estimation errors bθn θ 2. We repeat this process for Nruns different instances and for various choices of d. |
| Researcher Affiliation | Academia | 1School of Computer and Communication Sciences, EPFL, Lausanne, Vaud, Switzerland. Correspondence to: Marc Jourdan <EMAIL>. |
| Pseudocode | No | The paper describes methods and mathematical formulations (e.g., in sections '3. Preference-based M-estimator' and '4. Beyond M-estimators') but does not contain any explicit 'Pseudocode' or 'Algorithm' blocks, nor structured steps formatted like code. |
| Open Source Code | Yes | Reproducibility. Code for reproducing our empirical results is available at https://github.com/tml-epfl/ learning-parametric-distributions-from-samples-and-preferences. |
| Open Datasets | No | The paper conducts experiments based on generating data from "Gaussian distributions" (Section 6) and "Laplace and Rayleigh distributions" (Appendix H.1), specifying parameters like "uniformly drawn mean parameter θ U([1, 2]d)" and "isotropic covariance Σ = Id". It does not reference or provide access information for any publicly available or open datasets. |
| Dataset Splits | No | The paper describes generating synthetic data for experiments (e.g., "For sample size n [Nmax] with Nmax = 104, we compute the estimation errors bθn θ 2"). It mentions sample sizes but does not specify any training, testing, or validation dataset splits, as it generates data from parametric distributions rather than using pre-defined datasets. |
| Hardware Specification | Yes | Our experiments are conducted on 12 Intel(R) Core(TM) Ultra 7 165U 4.9GHz CPU. |
| Software Dependencies | No | The paper states: "Our code is implemented in Julia (Bezanson et al., 2017), version 1.11.5." While Julia has a specific version, other mentioned software like Stats Plots, Ju MP, Ipopt, and Hi GHS are listed without explicit version numbers within the text, failing to provide specific version numbers for multiple key components or the solvers themselves. |
| Experiment Setup | Yes | The paper provides details for its experimental setup, including the parameters for synthetic data generation: "we consider a uniformly drawn mean parameter θ U([1, 2]d) and the isotropic covariance Σ = Id." It also specifies the sample size and number of runs: "For sample size n [Nmax] with Nmax = 104, we compute the estimation errors bθn θ 2. We repeat this process for Nruns different instances and for various choices of d." and for Figure 2, "n = 10^4 and Nruns = 10^3". |