An Optimal Structured Zeroth-order Algorithm for Non-smooth Optimization

Authors: Marco Rando, Cesare Molinari, Lorenzo Rosasco, Silvia Villa

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conclude with numerical simulations where assumptions are satisfied, observing that our algorithm has very good practical performances.
Researcher Affiliation Academia Ma LGa-DIBRIS, University of Genova, IT (marco.rando@edu.unige.it, lorenzo.rosasco@unige.it). Ma LGa DIMA, University of Genova, Italy (molinari@dima.unige.it, silvia.villa@unige.it). Istituto Italiano di Tecnologia, Genova, Italy and CBMM MIT, Cambridge, MA, USA
Pseudocode Yes Algorithm 1 O-ZD: Orthogonal Zeroth-order Descent
Open Source Code No The paper mentions that scripts were implemented in Python and used specific libraries, but it does not provide an explicit statement about open-sourcing the code for the described methodology or a link to a repository.
Open Datasets No The paper uses 'synthetic target functions' (f1 and f2) and other defined functions (e.g., 'Sparse Group Lasso', 'Huber Loss') for its experiments. While the functions are defined, the paper does not provide concrete access information (link, DOI, repository, or standard citation for a publicly available dataset) for the data used in the experiments, as it seems to be generated synthetically for each run.
Dataset Splits No The paper describes experimental results but does not provide explicit details on dataset splits for training, validation, or testing (e.g., '80/10/10 split' or '5-fold cross-validation').
Hardware Specification Yes Table 1: Machine used to perform the experiments - OS Debian GNU/Linux 11 CPU(s) 4 x Intel(R) Core(TM) i7-1165G7 11th Gen @ 2.80GHz CPU Core(s) 4 RAM 8 GB
Software Dependencies Yes We implemented every script in Python3 (version 3.9.11) and used numpy (version 1.22.2) [27] and matplotlib (version 3.5.1) [29] libraries.
Experiment Setup Yes For these experiments, we consider d = 50 and we use, for the smooth convex case, the following parameters αk = 0.99 ℓ / (d L1) and hk = 10^-5. For the non-smooth target, we used αk = c * ℓ / (d * k^(1/2)) * 10^-5 and hk = 10^-7. ... For every method, we selected c = 0.65 except for the method with multiple Gaussian directions in which we selected c = 0.08.