Exploiting the Structure: Stochastic Gradient Methods Using Raw Clusters

Authors: Zeyuan Allen-Zhu, Yang Yuan, Karthik Sridharan

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments for three datasets that can be found on the Lib SVM website [8]: COVTYPE.BINARY, SENSIT (combined scale), and NEWS20.BINARY.
Researcher Affiliation Academia Zeyuan Allen-Zhu Princeton University / IAS zeyuan@csail.mit.edu Yang Yuan Cornell University yangyuan@cs.cornell.edu Karthik Sridharan Cornell University sridharan@cs.cornell.edu
Pseudocode Yes Algorithm 1 Cluster ACDM Algorithm 2 Cluster SVRG
Open Source Code No The paper mentions "The full version of this paper can be found on https://arxiv.org/abs/1602.02151," which is a link to the paper itself, not the source code for the described methodology. No other statements or links for code availability are provided.
Open Datasets Yes We conduct experiments for three datasets that can be found on the Lib SVM website [8]: COVTYPE.BINARY, SENSIT (combined scale), and NEWS20.BINARY. [8] Rong-En Fan and Chih-Jen Lin. LIBSVM Data: Classification, Regression and Multi-label. Accessed: 2015-06.
Dataset Splits No The paper mentions parameter tuning and cross-validation for choosing lambda, but does not provide specific details on the train/validation/test dataset splits used for their experiments.
Hardware Specification No The paper does not provide specific details regarding the hardware used for running the experiments (e.g., specific CPU/GPU models, memory).
Software Dependencies No The paper mentions using the "approximate nearest neighbor algorithm library E2LSH [2]" but does not provide a specific version number for this or any other software dependency. [2] points to http://www.mit.edu/ andoni/LSH/, 2004.
Experiment Setup Yes We use default epoch length m = 2n and Option I for SVRG. We use m = 2n and Option I for Cluster SVRG. ... For Lasso, ... We choose this dummy regularizer to have weight 10 7 for Covtype and Sense IT, and weight 10 6 for News20.