Scaling-Up Robust Gradient Descent Techniques

Authors: Matthew J. Holland7694-7701

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we study the efficiency and robustness of the proposed algorithm and its key competitors in a tightly controlled simulated setting (section ), verifying a substantial improvement in the cost-performance tradeoff, robustness to heavy-tailed data, and performance that scales well to higher dimensions.
Researcher Affiliation Academia Matthew J. Holland Osaka University matthew-h@ar.sanken.osaka-u.ac.jp
Pseudocode Yes Algorithm 1 Robust divide and conquer archetype; DC-SGD [Zn, w0; k].
Open Source Code Yes Repository: https://github.com/feedbackward/sgd-roboost
Open Datasets No The paper describes a simulated experimental setup ('we provide the learner with random losses of the form L(w; Z) = ( w w , X +E)2/2') rather than using a publicly available dataset with concrete access information.
Dataset Splits No The paper describes a simulated setting and does not provide specific train/validation/test dataset split information for a pre-existing dataset.
Hardware Specification No The paper states 'Complete details of the experimental setup are provided in the supplementary materials' but does not specify any hardware details like GPU/CPU models or specific machine configurations in the main text.
Software Dependencies No The paper states 'Complete details of the experimental setup are provided in the supplementary materials' but does not list any specific software dependencies with version numbers in the main text.
Experiment Setup No The paper states 'Complete details of the experimental setup are provided in the supplementary materials' and 'All detailed settings are in the supplementary materials.' Therefore, the main text does not contain specific experimental setup details like hyperparameters or training configurations.