Tensor Balancing on Statistical Manifold

Authors: Mahito Sugiyama, Hiroyuki Nakahara, Koji Tsuda

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental show in numerical experiments that the proposed algorithm is several orders of magnitude faster than existing ones.
Researcher Affiliation Academia 1National Institute of Informatics 2JST PRESTO 3RIKEN Brain Science Institute 4Graduate School of Frontier Sciences, The University of Tokyo 5RIKEN AIP 6NIMS.
Pseudocode No The paper describes the algorithms in prose and mathematical equations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes An implementation of algorithms for matrices and third order tensors is available at: https://github.com/ mahito-sugiyama/newton-balancing
Open Datasets Yes Hessenberg Matrix. The first set of experiments used a Hessenberg matrix, which has been a standard benchmark for matrix balancing (Parlett & Landis, 1982; Knight & Ruiz, 2013). Next, we collected a set of Trefethen matrices from a collection website2, which are nonnegative diagonal matrices with primes. Footnote 2: http://www.cise.ufl.edu/research/sparse/ matrices/
Dataset Splits No The paper evaluates algorithm efficiency and convergence on benchmark matrices but does not mention or specify any training, validation, or test dataset splits.
Hardware Specification Yes All experiments were conducted on Amazon Linux AMI release 2016.09 with a single core of 2.3 GHz Intel Xeon CPU E5-2686 v4 and 256 GB of memory.
Software Dependencies Yes All methods were implemented in C++ with the Eigen library and compiled with gcc 4.8.31.
Experiment Setup Yes We measured the residual of a matrix A = (a ij) by the squared norm (A 1 1, A T 1 1) 2, where each entry a ij is obtained as npij in our algorithm, and ran each of three algorithms until the residual is below the tolerance threshold 10 6.