A Box-Constrained Approach for Hard Permutation Problems

Authors: Cong Han Lim, Steve Wright

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We compare the objective values and timings achieved by FAQ and CD on problems from QAPLIB (Burkard et al., 1997), the canonical library of QAP problems, which has 137 problems with sizes varying from n = 10 to n = 256. Each method was run 100 times, and we compare the objective values corresponding to the best result and the median.
Researcher Affiliation Academia Cong Han Lim CONGHAN@CS.WISC.EDU Stephen J. Wright SWRIGHT@CS.WISC.EDU Department of Computer Sciences, University of Wisconsin-Madison, Madison, WI 53706, USA
Pseudocode Yes Algorithm 1 Continuation Algorithm for (12) and Algorithm 2 A cycle of cyclic coordinate descent for (12)
Open Source Code No We implemented the continuation method with the fast cyclic coordinate descent algorithm in MATLAB. While the paper mentions implementing their method in MATLAB, it does not provide any specific link or statement indicating that their own source code is publicly available.
Open Datasets Yes We compare the objective values and timings achieved by FAQ and CD on problems from QAPLIB (Burkard et al., 1997), the canonical library of QAP problems...
Dataset Splits No The paper focuses on solving quadratic assignment problems and combinatorial optimization. It discusses running the optimization algorithms on QAPLIB instances and synthetic instances, but it does not specify data splits like training, validation, or test sets as would be relevant for machine learning model training.
Hardware Specification No The paper states 'The experiments were run in MATLAB 8.4, limited to a single thread,' but it does not provide any specific hardware details such as CPU/GPU models, memory, or other computer specifications used for running the experiments.
Software Dependencies Yes The experiments were run in MATLAB 8.4, limited to a single thread.
Experiment Setup Yes We set ϵdec = 0.1% and ϵbin = 0.1 n in Algorithm 1. ... We decrease µ by Lr QAP/10 between subproblems (see (11) for the definition of Lr QAP), and terminate after solving the problem for two successive values of µ less than Lr QAP. In each run, we let the algorithm apply the new sorting network heuristic up to 3 times.