Efficient preconditioned stochastic gradient descent for estimation in latent variable models

Authors: Charlotte Baey, Maud Delattre, Estelle Kuhn, Jean-Benoist Leger, Sarah Lemler

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We illustrate through relevant simulations the performance of the proposed methodology in a nonlinear mixed effects model and in a stochastic block model.
Researcher Affiliation Academia 1Univ. Lille, CNRS, UMR 8524 Laboratoire Paul Painlev e, F-59000 Lille, France 2Universit e Paris-Saclay, INRAE, Ma IAGE, 78350, Jouy-en-Josas, France. 3Universit e de technologie de Compi egne, CNRS, Heudiasyc, Compi egne, France 4Universit e Paris-Saclay, Centrale Sup elec, Math ematiques et Informatique pour la Complexit e et les Syst emes, 91190, Gif-sur-Yvette, France.
Pseudocode Yes Algorithm 1 Fisher-SGD in the independent case
Open Source Code Yes The code is available in the Git repository https: //github.com/baeyc/fisher-sgd-nlme. [...] The code is available in the Git repository https://gitlab.com/jbleger/sbm_with_ fisher-sgd.
Open Datasets Yes We applied our algorithm to a real dataset from a study on coucal growth rates (Goymann et al., 2016).
Dataset Splits No The paper uses simulated datasets for evaluation and does not describe explicit train/validation/test splits of a larger dataset in the traditional machine learning sense for model training.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., CPU/GPU models, memory specifications).
Software Dependencies No The paper mentions the "R package saemix" and "Python module parametrization cookbook" but does not specify any version numbers for these or other software dependencies.
Experiment Setup Yes The authors propose to use Kpre-heating = 1000 and γ0 = 10 4. [...] Concerning the choice of α, to ensure Pk γk = + and Pk γ2 k < + , the authors propose to use α = 2/3. [...] The tuning parameters of the algorithm were set as follows: Kpre heating = 2000, K = 10000, Cheating = 100, α = 2/3, λ0 = 10 4, and the algorithm was initialized at a random value for θ.