Zeroth-Order Methods for Nonconvex Stochastic Problems with Decision-Dependent Distributions

Authors: Yuya Hikima, Akiko Takeda

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our simulation experiments with real data on a retail service application show that our methods output solutions with lower objective values than the conventional zeroth-order methods.
Researcher Affiliation Academia 1Graduate School of Information Science and Technology, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8654, Japan 2Center for Advanced Intelligence Project, RIKEN, 1-4-1 Nihonbashi, Chuo-ku, Tokyo 103-0027, Japan EMAIL, EMAIL
Pseudocode Yes Algorithm 1: Zeroth-order method with the improved one-point gradient estimator Algorithm 2: Zeroth-order method with the two-point gradient estimator
Open Source Code Yes Code https://github.com/Yuya-Hikima/AAAI25-Zeroth Order-Methods-for-Nonconvex-Stochastic-Problemswith-Decision-Dependent-Distributions
Open Datasets Yes We performed simulation experiments with real retail data from a supermarket service provider in Japan.5 All 5We used publicly available data, New Product Sales Ranking , provided by KSP-SP Co., Ltd, http://www.ksp-sp.com. Accessed August 15, 2024.
Dataset Splits No The paper mentions using real retail data but does not specify how this data was split into training, validation, or test sets for the experiments.
Hardware Specification Yes All experiments were conducted on a computer with an AMD EPYC 7413 24-Core Processor, 503.6 Gi B of RAM, and Ubuntu 20.04.6 LTS.
Software Dependencies Yes The program code was implemented in Python 3.8.3.
Experiment Setup Yes Proposed-1 (mini-batch). We implemented Algorithm 1 with µ0 := 0.19, µmin := 0.0001, c0 := P20 j=1 f(x0, ξj(x0)), smax := 10, β := 0.001 0.95k+1, γ = 0.95, mk = 30 + 2k, and M = 0.1, where k is the current iteration number.