Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Path-Specific Counterfactual Fairness via Dividend Correction
Authors: Daisuke Hatano, Satoshi Hara, Hiromi Arai
TMLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the effectiveness of the proposed algorithm by measuring the PSE, accuracy, and runtime using synthetic, Adult (Bache & Lichman, 2013), and German datasets. |
| Researcher Affiliation | Academia | Daisuke Hatano EMAIL RIKEN Center for Advanced Intelligence Project Satoshi Hara EMAIL The University of Electro-Communication Hiromi Arai EMAIL RIKEN Center for Advanced Intelligence Project |
| Pseudocode | Yes | Algorithm 1 Dividend Correction |
| Open Source Code | No | The paper refers to source code from other papers ( |
| Open Datasets | Yes | Our Adult setup was based on the configuration established by (Nabi & Shpitser, 2018; Chikahara et al., 2021), for which we used the source code provided on their Github repositories45. ... We leveraged the Adult dataset from their website8, which consists of 7 features extracted from the original Adult dataset, shown in the appendix. ... 8https://www.yongkaiwu.com/publication/ ... Our German 6 setting was based on (Chikahara et al., 2021). ... 6https://www.kaggle.com/datasets/uciml/german-credit |
| Dataset Splits | Yes | We generated 10 instances, each of which is sampled 6000 data points and separated them into 5000 as training data and 1000 as test data. ... To construct the datasets, 5000 records were selected for training and 1000 records for testing from a total of 65,123 records. ... To construct the dataset, we selected 900 records for training data and 100 records for testing from a total of 1000 records. |
| Hardware Specification | Yes | The experiments were performed using a Macbook Pro with Apple M1 Max and 64GB RAM. |
| Software Dependencies | No | The paper mentions machine learning models and algorithms (e.g., logistic regression, random forest, neural network, EM algorithm) but does not provide specific software library names with version numbers. |
| Experiment Setup | Yes | For Algorithm 1, we proceeded line 6, where ϵ alters from 0 to 1 in increments of 0.05. For both FIO and PIU, their penalty parameters were modified with respect to fairness from 0 to 2 in steps of 0.1. ... As for CF, we altered a fairness parameter τ from 0.01 to 0.3 in steps of 0.01. ... We set T = 100 for the number of iterations of the EM algorithm. |