Revisiting Projection-Free Optimization for Strongly Convex Constraint Sets

Authors: Jarrid Rector-Brooks, Jun-Kun Wang, Barzan Mozafari1576-1583

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 7 Experiments We have conducted extensive experiments on different combinations of loss functions, constraint sets, and real-life datasets (Table 2). Here, we only report two main sets of experiments: the empirical validation of our theoretical results in terms of convergence rates (Section 7.1) and the comparison of various optimizations in terms of actual run times (Section 7.2).
Researcher Affiliation Academia Jarrid Rector-Brooks 2260 Hayward St Ann Arbor, MI, 48104 University of Michigan, Ann Arbor jrectorb@umich.edu Jun-Kun Wang* 226 Ferst Drive NW Atlanta, GA, 30332 Georgia Institute of Technology jimwang@gatech.edu Barzan Mozafari 2260 Hayward St Ann Arbor, MI, 48104 University of Michigan, Ann Arbor mozafari@umich.edu
Pseudocode Yes Algorithm 1 Standard Frank-Wolfe algorithm
Open Source Code No Our plan is to integrate PA in machine learning libraries libraries, including our Blink ML project (Park et al. 2018).
Open Datasets Yes For regression, we used the Year Prediction MSD dataset (500K observations, 90 features) (Lichman 2013). For classification, we used the Adult dataset (49K observations, 14 features) (Lichman 2013). For matrix completion, we used the Movie Lens dataset (1M movie ratings from 6,040 users on 3,900 movies) (Harper and Konstan 2016).
Dataset Splits No The paper mentions using specific datasets but does not provide details on how these datasets were split into training, validation, or test sets.
Hardware Specification No The paper does not provide specific details about the hardware used for running its experiments.
Software Dependencies No The paper does not provide specific details about ancillary software dependencies, including their version numbers, that would be needed to replicate the experiments.
Experiment Setup No The paper describes the algorithms used and the tasks/datasets, but it does not provide specific experimental setup details such as hyperparameter values (e.g., learning rates, batch sizes, number of epochs) or other system-level training configurations for the empirical evaluations.