Decentralized Gradient-Free Methods for Stochastic Non-smooth Non-convex Optimization

Authors: Zhenwei Lin, Jingfan Xia, Qi Deng, Luo Luo

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

Reproducibility Variable Result LLM Response
Research Type Experimental Moreover, experimental results underscore the empirical advantages of our proposed algorithms when applied to real-world datasets.
Researcher Affiliation Academia 1School of Information Management and Engineering, Shanghai University of Finance and Economics 2School of Data Science, Fudan University
Pseudocode Yes Algorithm 1: DGFM at each node i
Open Source Code No No explicit statement about providing open-source code for the methodology was found.
Open Datasets Yes Data: We evaluate our proposed algorithms using several standard datasets in LIBSVM (Chang and Lin 2011), which are described in Table 1.
Dataset Splits No The paper mentions training and testing data but does not explicitly provide details about validation dataset splits (percentages, counts, or explicit standard validation sets).
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory, cloud instance types) used for running experiments were provided.
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries used in the experiments.
Experiment Setup Yes Throughout all the experiments, we set δ = 0.001 and tune the stepsize η from {0.0005, 0.001, 0.005, 0.01} for all four algorithms and b from {10, 100, 500}, T from {10, 50, 100} for DGFM+ and GFM+, T from {1, 5, 10} for DGFM+, m = 20 for two decentralized algorithms. ... Throughout all the experiments, we set δ = 0.01, b = {16, 32, 64}. For DGFM+ and GFM+, we tune b from {40, 80, 800, 1600}, T from {2, 5, 10, 20}. Additionally, tune T from {1, 10, 20} for DGFM+. For all algorithms, we tune the stepsize η from {0.05, 0.1, 0.5, 1} and multiply a decay factor 0.6 if no improvement in 300 iterations. For all experiments, we set the initial perturbation as 0.