Debiasing Averaged Stochastic Gradient Descent to handle missing values

Authors: Aude Sportisse, Claire Boyer, Aymeric Dieuleveut, Julie Josse

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show the convergence behavior and the relevance of the algorithm not only on synthetic data but also on real data sets, including those collected from medical register. and 5 Experiments, 5.1 Synthetic data, 5.2 Real dataset 1: Traumabase dataset, 5.3 Real dataset 2: Superconductivity dataset
Researcher Affiliation Academia Aude Sportisse Sorbonne University Paris, France, Claire Boyer Sorbonne University Paris, France, Aymeric Dieuleveut Ecole Polytechnique Palaiseau, France, Julie Josse INRIA Montpellier, France
Pseudocode Yes Algorithm 1 Averaged SGD for Heterogeneous Missing Data
Open Source Code Yes The code to reproduce all the simulations and numerical experiments is available on https://github.com/Aude Sportisse/SGD-NA.
Open Datasets No The paper mentions 'Traumabase dataset' and 'Superconductivity dataset (available here)' but does not provide a direct URL, DOI, or specific citation with authors and year for public access within the provided text.
Dataset Splits No The dataset is divided into training and test sets (random selection of 70 30%). No explicit mention of a validation set split.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, processor types, or memory specifications used for running the experiments.
Software Dependencies No The paper mentions 'sklearn.impute.IterativeImputer' but does not provide a specific version number, nor does it list other software dependencies with their versions.
Experiment Setup Yes Av SGD described in Algorithm 1 with a constant step size α = 1 2L, and L given in (6). and The regularization parameter λ (see Remark 1) is chosen by cross validation.