Approximating Orthogonal Matrices with Effective Givens Factorization

Authors: Thomas Frerix, Joan Bruna

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate numerical results of approximating the graph Fourier transform...
Researcher Affiliation Academia 1Technical University of Munich 2New York University.
Pseudocode Yes Algorithm 1 Coordinate descent on the L1-criterion
Open Source Code Yes An implementation of these algorithms can be found at https://github.com/tfrerix/ givens-factorization
Open Datasets Yes MINNESOTA 2642 3304 (Defferrard et al.) HUMANPROTEIN 3133 6726 (Rual et al., 2005) EMAIL 1133 5451 (Guimer a et al., 2003) FACEBOOK 2888 2981 (Mc Auley & Leskovec, 2012)
Dataset Splits No The paper describes the generation of datasets (planted models, Barabasi-Albert graphs, and uses real-world graph datasets) but does not specify any train/validation/test splits, percentages, or explicit methodologies for partitioning data.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU, GPU models, memory, cloud instances) used for running the experiments.
Software Dependencies No While an implementation link is provided, the paper does not explicitly list software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x) within its text.
Experiment Setup Yes To obtain a Givens sequence, we factorize these samples with manifold coordinate descent on the L1-objective (13). Along the optimization path, we define Nϵ(U) as the number of Givens factors for which the normalized approximation error (3) is smaller than ϵ = 0.1, i.e., Nϵ(U) := min N ||U − G1 . . . GN ||F,sym < ϵ