Sparse Quadratic Optimisation over the Stiefel Manifold with Application to Permutation Synchronisation

Authors: Florian Bernard, Daniel Cremers, Johan Thunberg

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally compare our proposed approach with various methods for permutation synchronisation and perform an evaluation on both real and synthetic datasets.
Researcher Affiliation Academia Florian Bernard TU Munich, University of Bonn Daniel Cremers TU Munich Johan Thunberg Halmstad University
Pseudocode Yes Algorithm 1: Overview of our proposed algorithm.
Open Source Code No The paper does not contain an explicit statement or link indicating that the source code for the methodology described is publicly available or released by the authors.
Open Datasets Yes In this experiment we use the CMU house image sequence [1] comprising 111 frames within the experimental protocol of [34]. We reproduce the procedure described in [8] for generating synthetic instances for the synchronisation of partial permutations.
Dataset Splits No The paper describes generating and sampling problem instances but does not provide explicit details about train/validation/test splits, specific percentages, or cross-validation strategies typically used for model validation.
Hardware Specification Yes All experiments are run on a Macbook Pro (2.8 GHz quad core i7, 16 GB RAM), where for ϵ = 10 5 we use f(Ut)/f(Ut+1) 1 ϵ as convergence criterion in Algorithm 1, and a step size of αt = h(Ut, I) h T (Ut, I) 1 in (7).
Software Dependencies No The paper mentions using "the efficient implementation from the authors of [51]" and an "efficient implementation in [10]", but does not specify software dependencies (e.g., programming languages, libraries, frameworks) with version numbers for their own code.
Experiment Setup Yes All experiments are run on a Macbook Pro (2.8 GHz quad core i7, 16 GB RAM), where for ϵ = 10 5 we use f(Ut)/f(Ut+1) 1 ϵ as convergence criterion in Algorithm 1, and a step size of αt = h(Ut, I) h T (Ut, I) 1 in (7).