Optimal Black-Box Reductions Between Optimization Objectives

Authors: Zeyuan Allen-Zhu, Elad Hazan

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

Reproducibility Variable Result LLM Response
Research Type Experimental We perform experiments to confirm our theoretical speed-ups obtained for Adapt Smooth and Adapt Reg. We work on minimizing Lasso and SVM objectives for the following three well-known datasets that can be found on the Lib SVM website [10]: covtype, mnist, and rcv1.
Researcher Affiliation Academia Zeyuan Allen-Zhu zeyuan@csail.mit.edu Institute for Advanced Study & Princeton University Elad Hazan ehazan@cs.princeton.edu Princeton University
Pseudocode Yes Algorithm 1 The Adapt Reg Reduction Input: an objective F( ) in Case 2 (smooth and not necessarily strongly convex); x0 a starting vector, σ0 an initial regularization parameter, T the number of epochs; an algorithm A that solves Case 1 of problem (1.1). Output: bx T .
Open Source Code No The paper does not contain any explicit statement about making the source code available or provide a link to a code repository.
Open Datasets Yes We work on minimizing Lasso and SVM objectives for the following three well-known datasets that can be found on the Lib SVM website [10]: covtype, mnist, and rcv1. [10] Rong-En Fan and Chih-Jen Lin. LIBSVM Data: Classification, Regression and Multi-label. Accessed: 2015-06.
Dataset Splits No The paper mentions deferring 'dataset and implementation details' to the full version but does not provide specific training/test/validation dataset splits (e.g., percentages, sample counts, or citations to predefined splits) in the provided text.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper refers to methods like APCG [20] and SVRG [14] and mentions using datasets from the Lib SVM website [10], but it does not provide specific version numbers for any software components, libraries, or solvers used in the experiments.
Experiment Setup No The paper describes some practical implementation details such as termination criteria for the oracle in the inner loop (e.g., duality gap below 1/4 or Euclidean norm below 1/3 of the previous epoch), and ranges for regularization weights (e.g., {10k, 3 · 10k : k ∈ Z}). However, it does not provide specific hyperparameter values like learning rates, batch sizes, or optimizer settings for the experiments.