Boosting Frank-Wolfe by Chasing Gradients
Authors: Cyrille Combettes, Sebastian Pokutta
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compared the Boosted Frank-Wolfe algorithm (Boost FW, Algorithm 2) to the Away-Step Frank-Wolfe algorithm (AFW) (Wolfe, 1970), the Decomposition-Invariant Pairwise Conditional Gradient algorithm (DICG) (Garber & Meshi, 2016), and the Blended Conditional Gradients algorithm (BCG) (Braun et al., 2019) in a series of computational experiments. |
| Researcher Affiliation | Academia | 1School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, GA, USA 2Institute of Mathematics, Technische Universit at Berlin, Berlin, Germany 3Department for AI in Society, Science, and Technology, Zuse Institute Berlin, Berlin, Germany. Correspondence to: Cyrille W. Combettes <cyrille@gatech.edu>. |
| Pseudocode | Yes | Algorithm 1 Frank-Wolfe (FW) ... Algorithm 2 Boosted Frank-Wolfe (Boost FW) |
| Open Source Code | Yes | Code is available at https://github.com/cyrillewcombettes/boostfw. |
| Open Datasets | Yes | We consider the task of recognizing the handwritten digits 4 and 9 from the Gisette dataset (Guyon et al., 2005), available at https://archive.ics.uci.edu/ml/ datasets/Gisette. ... We consider the task of collaborative filtering on the Movie Lens 100k dataset (Harper & Konstan, 2015), available at https://grouplens.org/datasets/ movielens/100k/. ... You Tube-Objects dataset (Prest et al., 2012)... We obtained the data from https://github. com/Simon-Lacoste-Julien/linear FW. |
| Dataset Splits | No | The paper mentions using standard datasets like Gisette and MovieLens 100k but does not explicitly provide details on how the data was split into training, validation, and test sets, nor does it refer to a specific predefined split methodology for these experiments. |
| Hardware Specification | Yes | We ran the experiments on a laptop under Linux Ubuntu 18.04 with Intel Core i7 3.5GHz CPU and 8GB RAM. |
| Software Dependencies | No | The paper mentions using 'Python', 'networkx', and 'scipy.sparse.linalg' but does not provide specific version numbers for these software dependencies. |
| Experiment Setup | Yes | Unless specified otherwise, we set δ 10 3 and K + in Boost FW. We set m = 200, n = 500, σ = 0.05, and τ = x 1. We used m = 2000 samples and the number of features is n = 5000. We set τ = 10, L = 0.5, and δ 10 4 in Boost FW. |