Online Convex Optimization Over Erdos-Renyi Random Networks

Authors: Jinlong Lei, Peng Yi, Yiguang Hong, Jie Chen, Guodong Shi

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical studies have validated the theoretical findings.
Researcher Affiliation Academia Jinlong Lei Tongji University Shanghai, China leijinlong@tongji.edu.cn Peng Yi Tongji University Shanghai, China yipeng@tongji.edu.cn Yiguang Hong Tongji University Shanghai, China yghong@iss.ac.cn Jie Chen Tongji University Shanghai, China chenjie@bit.edu.cn Guodong Shi, The University of Sydney NSW, Australia guodong.shi@sydney.edu.au
Pseudocode Yes Algorithm 1 Distributed online gradient descent with full gradients; Algorithm 2 Distributed online algorithm with one-point bandit feedback; Algorithm 3 Distributed algorithm with two-points bandit feedback.
Open Source Code No The paper does not provide any concrete access to source code for the methodology described.
Open Datasets Yes we examine the empirical performance of the proposed distributed algorithms on the bodyfat dataset with 14 features and 252 instances from LIBSVM library 2. The data set is from https://www.csie.ntu.edu.tw/ cjlin/libsvmtools/datasets/
Dataset Splits No The paper uses the 'bodyfat dataset' but does not specify any training, validation, or test splits by percentages, sample counts, or predefined methods.
Hardware Specification No The paper does not provide any specific hardware details (e.g., CPU/GPU models, memory) used for running its experiments.
Software Dependencies No The paper does not provide any specific ancillary software details with version numbers (e.g., Python 3.8, PyTorch 1.9).
Experiment Setup Yes We set N = 30, p = 0.2, and µ = 0 in (11) to get convex losses. In addition, we set µ = 1 in (11) to construct strongly convex losses. We fix the time horizon T = 200. Let the link connection probability p vary from p = 0.1 to p = 0.9 at a stride of 0.1. We set N = 20, let the vector dimension d vary from d = 5 to d = 100.