Balancing Bias in Two-sided Markets for Fair Stable Matchings

Authors: Siyuan Wu, Leong Hou U, Panagiotis Karras

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our extensive experimental study shows that ISORROPIA surpasses the time-efficiency of baselines that return the exact solution by up to three orders of magnitude. We use synthetic and real datasets in our assessment, as follows: (i) Following a prior work (Tziavelis et al., 2020), we construct a dataset, Uniform, with preference lists drawn from the uniform distribution. (ii) We use the settings in (Siala & O Sullivan, 2017) to generate Hard instances by the method outlined in (Irving & Leather, 1986)... Figure 4 presents our results on runtime and balance cost.
Researcher Affiliation Academia Siyuan Wu University of Macau EMAIL; Leong Hou U University of Macau EMAIL; Panagiotis Karras U. of Copenhagen & Aarhus U. EMAIL
Pseudocode Yes Algorithm 1 ISORROPIA; Algorithm 2 ENUM; Algorithm 3 Local Search Strategies
Open Source Code Yes All methods were implemented in C++; the code is available in our Github repository.9 (Footnote 9: https://github.com/Asuka54089/Isorropia)
Open Datasets Yes We use synthetic and real datasets in our assessment, as follows: ... (iii) Taxi reflects the two-sided market of taxis and users, drawn from the NYC Taxi dataset6; ... (iv) Adm captures a two-sided market of university admissions; we employ university rankings7 and GRE and TOEFL scores to define8 preferences on two sides. (Footnote 6: https://www.nyc.gov/site/tlc/about/data.page; Footnote 7: https://kaggle.com/datasets/mylesoneill/world-university-rankings; Footnote 8: https://kaggle.com/datasets/akshaydattatraykhare/data-for-admission-in-the-university)
Dataset Splits No The paper describes the generation of some synthetic datasets and uses external real-world datasets, but it does not specify how these datasets were split into training, validation, or test sets for the experiments. It provides characteristics of the datasets in Table 6 but no split methodology.
Hardware Specification Yes We ran experiments on an Intel i5-13500H machine @2.60 GHz with 32G memory running Windows.
Software Dependencies No All methods were implemented in C++; the code is available in our Github repository. While C++ is mentioned, no specific version number for C++ or any other libraries/packages is provided.
Experiment Setup Yes (2) BILS (Viet et al., 2016b;a), a greedy local search method on the stable marriage lattice; we set the probability of random movement to p = 0.05; (4)POWERBALANCE (Tziavelis et al., 2019), a heuristic that goes through a series of proposal iterations from both sides by strongly deferred acceptance, whereby unmatched agents only accept proposals more preferable than their own target, with the maximum number of proposal rounds fixed to t = n log2 2 n/10; (5) HMS (Tziavelis et al., 2019), a heuristic that improves upon the results of POWERBALANCE by an m-step local search over k rounds, with complexity O(tn + kmn2).