Stochastic Subspace Cubic Newton Method

Authors: Filip Hanzely, Nikita Doikov, Yurii Nesterov, Peter Richtarik

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our numerical experiments show that SSCN outperforms non-accelerated first-order CD algorithms while being competitive to their accelerated variants. 8. Experiments We now numerically verify our theoretical claims.
Researcher Affiliation Academia 1King Abdullah University of Science and Technology, Thuwal, Saudi Arabia 2Catholic University of Louvain, Louvain-la-Neuve, Belgium.
Pseudocode Yes Algorithm 1 SSCN: Stochastic Subspace Cubic Newton 1: Initialization: x0, distribution D of random matrices with d rows and full column rank 2: for k = 0, 1, . . . do 3: Sample S from distribution D 4: hk = argminh Rτ(S) TS(xk, h) 5: Set xk+1 = xk + Shk
Open Source Code No The paper does not provide any explicit statement or link for the open-source code of the described methodology.
Open Datasets Yes We consider binary classification with LIBSVM (Chang & Lin, 2011) data modelled by regularized logistic regression.
Dataset Splits No The paper mentions using LIBSVM data but does not explicitly provide training, validation, or test dataset split percentages or counts in the main text provided.
Hardware Specification No The paper discusses computational cost in terms of operations (e.g., O(n), O(nτ^2 + τ^3)) but does not provide specific hardware details such as GPU/CPU models, memory, or processing units used for experiments.
Software Dependencies No The paper mentions 'LIBSVM' as data source and references various methods and frameworks (e.g., FGM) but does not list specific software dependencies with version numbers used for implementing their method or running experiments.
Experiment Setup No The paper states 'the exact setup for this experiment can be found in Section B of the Appendix' but does not provide specific experimental setup details such as hyperparameter values (e.g., learning rate, batch size, number of epochs) in the main text.