On the Second-order Convergence Properties of Random Search Methods

Authors: Aurelien Lucchi, Antonio Orvieto, Adamos Solomou

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our algorithm empirically and find good agreements with our theoretical results.
Researcher Affiliation Academia Aurelien Lucchi Antonio Orvieto Adamos Solomou Department of Computer Science ETH Zurich
Pseudocode Yes Algorithm 1 TWO-STEP RANDOM SEARCH (RS). Similar to the STP method [6], but we alternate between two perturbation magnitudes: σ1 is set to be optimal for the large gradient case, while σ2 optimal to escape saddles.
Open Source Code Yes the code for reproducing the experiments is available online5.
Open Datasets No The paper uses synthetic functions (e.g., 'Function with growing dimension' and 'Rastrigin function') for its experiments, which are generated and not referenced as publicly available datasets with access information.
Dataset Splits No The paper uses synthetic functions for optimization tasks and does not specify training, validation, or test dataset splits in the conventional sense.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies used in the experiments.
Experiment Setup Yes For each task, the hyperparameters of every method are selected based on a coarse grid search refined by trial and error. We choose to run DFPI for 20 iterations for all the results shown in the paper.