Sample-and-Bound for Non-convex Optimization

Authors: Yaoguang Zhai, Zhizhen Qin, Sicun Gao

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the proposed algorithms on high-dimensional nonconvex optimization benchmarks against competitive baselines and analyze the effects of the hyper parameters.
Researcher Affiliation Academia University of California, San Diego yazhai@ucsd.edu, zhizhenqin@ucsd.edu, sicung@ucsd.edu
Pseudocode Yes The pseudocode of MCIR is provided in Alg. 1
Open Source Code No No explicit statement or link to open-source code for the described methodology was found.
Open Datasets Yes To evaluate the performance of our algorithms, our benchmark sets include three distinct categories: synthetic functions designed for nonlinear optimization, bound-constrained non-convex global optimization problems derived from real-world scenarios, and neural networks fitted for single valued functions. [...] Synthetic functions are widely-used in nonlinear optimization benchmarks (Lavezzi, Guye, and Ciarci a 2022). These functions usually have numerous local minima, valleys, and ridges in their landscapes which is hard for normal optimization algorithms. In our tests, we choose three functions: Levy, Ackley, and Michalewicz [...] For our evaluation of non-convex global optimization problems in various fields, we select bound-constrained problems from the collection presented in (The Optimization Firm 2023; Puranik and Sahinidis 2017) that do not involve any additional inequality or equality constraints.
Dataset Splits No The paper uses benchmark functions but does not explicitly provide specific train/validation/test dataset splits, percentages, or sample counts.
Hardware Specification Yes We conduct our experiments on a local machine with Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz, 16G RAM, and NVIDIA Ge Force GTX 1080 graphic card.
Software Dependencies Yes Gurobi (Gurobi Optimization 2023) is a widely used commercial optimization solver [...] CMA-ES/pycma: r3.3.0
Experiment Setup Yes In this formula, Clb, Cv and Cx are weights for the function s lower bound, the volume of the box, and visitationbased exploration, respectively. [...] In most cases we cap the number of iterations at fewer than 50, as we do not want to overemphasize the choice of the local optimizer.