Adaptive Newton Sketch: Linear-time Optimization with Quadratic Convergence and Effective Hessian Dimensionality

Authors: Jonathan Lacotte, Yifei Wang, Mert Pilanci

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

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we show in Section 5 the empirical benefits of our adaptive method, compared to several standard optimization baselines. and In this section, we compare adaptive Newton Sketch (NSada) with other optimization methods in regularized logistic regression problems as in Example 4. The datasets used in the numerical experiments are collected from LIBSVM (Chih-Chung & Chih-Jen, 2011). For each dataset, we randomly split it into a training set and a test set with the ratio 1 : 1.
Researcher Affiliation Academia Jonathan Lacotte 1 Yifei Wang 1 Mert Pilanci 1 1Department of Electrical Engineering, Stanford University.
Pseudocode Yes Algorithm 1 Effective dimension Newton sketch and Algorithm 2 Adaptive effective dimension Newton sketch
Open Source Code No The paper does not include an explicit statement about releasing source code for the described methodology or a link to a code repository.
Open Datasets Yes The datasets used in the numerical experiments are collected from LIBSVM (Chih-Chung & Chih-Jen, 2011).
Dataset Splits Yes For each dataset, we randomly split it into a training set and a test set with the ratio 1 : 1.
Hardware Specification Yes All numerical experiments are executed on a Dell Power Edge R840 workstation. Specifically, we use 4 cores with 192GB ram for all compared methods.
Software Dependencies No The paper mentions software like 'Katyusha' and 'SVRG' and references LIBSVM, but does not provide specific version numbers for any software libraries or dependencies used in the experiments.
Experiment Setup Yes For SVRG and Katyusha, we use a batch size of 20.