Faster Gradient-Free Methods for Escaping Saddle Points

Authors: Hualin Zhang, Bin Gu

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we conduct several numerical experiments to verify the effectiveness of the proposed methods for escaping saddle points and the efficiency compared with the existing methods.
Researcher Affiliation Academia Hualin Zhang 1, Bin Gu 1,2 1Nanjing University of Information Science & Technology 2MBZUAI {zhanghualin98, jsgubin}@gmail.com
Pseudocode Yes Algorithm 1 Zeroth-Order Perturbed Accelerated Gradient Descent, Algorithm 2 Negative Curvature Exploitation (xt, vt, s), Algorithm 3 Zeroth-Order Perturbed Accelerated Gradient Descent with Accelerated Negative Curvature Finding, Algorithm 4 Zeroth-Order Accelerated Negative Curvature Finding without Renormalization( x, r , T )
Open Source Code No The paper does not provide an explicit statement or link for open-source code related to the methodology described.
Open Datasets No The paper evaluates its methods on a 'cubic regularization problem' and a 'quartic function', both of which are mathematical formulations or synthetic problems, not publicly available datasets with specific access information.
Dataset Splits No The paper does not provide specific details on training, validation, or test dataset splits. It appears to use synthetic functions initialized from a saddle point.
Hardware Specification Yes All experiments are performed on a computer with a six-core Intel Core i5-10500 CPU.
Software Dependencies No The paper does not provide specific version numbers for its software dependencies or libraries used in the implementation.
Experiment Setup Yes In this experiment, we set ϵ = 10 2. ... For Algorithm 1 and 3, the parameter settings basically follow Eq. (5) and Eq. (7). Specifically, we choose ϵ = 0.001 and the perturbation radius r and r are set to 0.001. The Lipschitz constants ℓand ρ are selected based on a coarse grid search of the region {0.1, 1, 10, 100} {0.1, 1, 10, 100}. Table 2: Parameter settings of the cubic regularization problem experiment. Table 3: Parameter settings of the cubic quartic function experiment.