Smooth Bilevel Programming for Sparse Regularization

Authors: Clarice Poon, Gabriel Peyré

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform a numerical benchmark of the convergence speed of our algorithm against state of the art solvers for Lasso, group Lasso, trace norm and linearly constrained problems. These results highlight the versatility of our approach, removing the need to use different solvers depending on the specificity of the ML problem under study.
Researcher Affiliation Academia Department of mathematical sciences, University of Bath, Bath BA2 7AY, UK cmshp20@bath.ac.uk CNRS and DMA, Ecole Normale Supérieure, PSL University, 45 rue d Ulm, F-75230 PARIS cedex 05, FRANCE, gabriel.peyre@ens.fr
Pseudocode No The paper describes algorithms but does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes The code to reproduce the results of this article is available online3. 3 https://github.com/gpeyre/2021-Non Cvx Pro
Open Datasets Yes We tested on 8 datasets from the Libsvm repository4. 4 https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/
Dataset Splits Yes Here λ 0 is the regularisation parameter which is typically tuned by cross-validation, and in the limit case λ = 0, (P0) is a constraint problem minβ R(β) under the constraint L(Xβ, y) = 0.
Hardware Specification Yes All numerics are conducted on 2.4 GHz Quad-Core Intel Core i5 processor with 16GB RAM.
Software Dependencies No The paper mentions software like L-BFGS, FISTA, SPG/Spa RSA, CELER, etc., but does not provide specific version numbers for any of them.
Experiment Setup No The paper describes the choice of regularization parameters and general solution methods (L-BFGS, Cholesky solver) but does not provide specific hyperparameter values (e.g., learning rates, batch sizes, number of epochs, or detailed optimizer settings) for the experiments.