Guided Zeroth-Order Methods for Stochastic Non-convex Problems with Decision-Dependent Distributions

Authors: Yuya Hikima, Hiroshi Sawada, Akinori Fujino

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our simulation experiments on multiple applications show that our methods output solutions with lower objective values than the existing zeroth-order methods do.
Researcher Affiliation Industry 1Communication Science Laboratories, NTT Corporation, Kyoto, Japan. Correspondence to: Yuya Hikima <EMAIL>.
Pseudocode Yes Algorithm 1 Guided zeroth-order method with new samples Algorithm 2 Guided zeroth-order method with historical samples
Open Source Code No The program code was implemented in Python 3.12.2. The paper does not contain a specific link to a code repository or an explicit statement about open-sourcing the code for the methodology described.
Open Datasets Yes The data, New Product Sales Ranking , has been made publicly available by KSP-SP Co., Ltd, http://www.ksp-sp.com. Last accessed on January 28, 2025. We conducted experiments on the application of strategic classification with a real dataset from (Yeh & hui Lien, 2009).7 As with (Levanon & Rosenfeld, 2021), we used a preprocessed version of the data by (Ustun et al., 2019).
Dataset Splits Yes We divided 13272 data samples into a 12272-sample training set and 1000-sample test set in our experiments.
Hardware Specification Yes All experiments were conducted on a computer with Intel(R) Xeon(R) CPU E5-2697A v4 (2.60GHz) x2 and 512GB of memory RAM.
Software Dependencies Yes The program code was implemented in Python 3.12.2.
Experiment Setup Yes GZO-NS. This is Algorithm 1 with µ0 := 0.2, µmin := 0.0001, α0 = 0, βk := 0.01 0.95k, η = 0.95, γ = 0.98, mk := 30 + 2k, and nk := 30 + 2k, where k is the current iteration number. Details of the parameters can be found in Appendix A.1.2 and A.2.1.