Distributed Zero-Order Optimization under Adversarial Noise

Authors: Arya Akhavan, Massimiliano Pontil, Alexandre Tsybakov

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

Reproducibility Variable Result LLM Response
Research Type Experimental In Appendix E, we present a numerical comparison between the algorithm proposed in this paper and that in Akhavan et al. [2020]. The results confirm our theoretical findings. The algorithm of this paper converges faster and the advantage is more pronounced as d increases.
Researcher Affiliation Academia Arya Akhavan CSML, Istituto Italiano di Tecnologia and CREST, ENSAE, IP Paris aria.akhavanfoomani@iit.it Massimiliano Pontil CSML, Istituto Italiano di Tecnologia and University College London massimiliano.pontil@iit.it Alexandre B. Tsybakov CREST, ENSAE, IP Paris alexandre.tsybakov@ensae.fr
Pseudocode Yes Algorithm 1 Distributed Zero-Order Gradient [...] Algorithm 2 Gradient Estimator with 2d Queries
Open Source Code No The paper does not provide any statement or link indicating that the source code for their proposed method is open-source or publicly available.
Open Datasets No The paper describes theoretical algorithms and their convergence properties, and mentions a numerical comparison, but does not specify or provide access information for any publicly available or open dataset used in its evaluation.
Dataset Splits No The paper does not specify any dataset splits (e.g., training, validation, test percentages or counts) needed to reproduce experiments, as it does not describe specific experiments on datasets in detail.
Hardware Specification No The paper does not provide any specific hardware details (e.g., GPU/CPU models, memory) used for running its numerical comparisons or experiments.
Software Dependencies No The paper does not list specific software components with version numbers (e.g., Python, PyTorch, TensorFlow, or specific solvers) required to reproduce the work.
Experiment Setup No The paper specifies mathematical tuning parameters for the algorithm (ηt and ht) but does not provide concrete experimental setup details such as hyperparameters, learning rates, batch sizes, or optimizer settings for a numerical evaluation.