Distributionally Robust Parametric Maximum Likelihood Estimation

Authors: Viet Anh Nguyen, Xuhui Zhang, Jose Blanchet, Angelos Georghiou

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

Reproducibility Variable Result LLM Response
Research Type Experimental We now showcase the abilities of the proposed framework in the distributionally robust Poisson and logistic regression settings using a combination of simulated and empirical experiments. 5 Numerical Experiments
Researcher Affiliation Academia Viet Anh Nguyen Xuhui Zhang Jos e Blanchet Stanford University, United States {viet-anh.nguyen, xuhui.zhang, jose.blanchet}@stanford.edu Angelos Georghiou University of Cyprus, Cyprus georghiou.angelos@ucy.ac.cy
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The MATLAB code is available at https://github.com/angelosgeorghiou/DR-Parametric-MLE.
Open Datasets Yes using data sets from the UCI repository [13].
Dataset Splits Yes In each independent trial, we randomly split the data into train-validation-test set with proportion 50%-25%-25%.
Hardware Specification Yes on an Intel i7 CPU (1.90GHz) computer.
Software Dependencies Yes modeled in MATLAB using CVX [16] and solved by the exponential conic solver MOSEK [24]. (Referencing [16] 'CVX: Matlab software for disciplined convex programming, version 2.1, Mar. 2014.' and [24] 'MOSEK Ap S. The MOSEK optimization toolbox. Version 9.2., 2019.')
Experiment Setup Yes We calibrate the regression model (12) by tuning ρc = a N 1 c with a [10 4, 1] and ε [PC c=1 bpcρc, 1], both using a logarithmic scale with 20 discrete points. we calibrate the regression model (13) by tuning ρc = a N 1 c with a [10 4, 10] using a logarithmic scale with 10 discrete points and setting ε = 2 PC c=1 bpcρc.