Interventional Fairness on Partially Known Causal Graphs: A Constrained Optimization Approach

Authors: Aoqi Zuo, Yiqing Li, Susan Wei, Mingming Gong

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Results on both simulated and real-world datasets demonstrate the effectiveness of this method.
Researcher Affiliation Academia Aoqi Zuo1, Yiqing Li1,2,3, Susan Wei1 & Mingming Gong1,3 1School of Mathematics and Statistics, The University of Melbourne 2School of Data Science, Fudan University 3Mohamed bin Zayed University of Artificial Intelligence azuo@student.unimelb.edu.au yiqingli20@fudan.edu.cn {susan.wei,mingming.gong}@unimelb.edu.au
Pseudocode Yes Algorithm 1 Construct MPDAG (Perkovic et al., 2017; Meek, 1995) and Algorithm 2 Partial causal ordering (PCO) (Perkovic, 2020, Algorithm 1)
Open Source Code No The paper does not explicitly state that open-source code for the methodology is provided, nor does it include a link to a code repository.
Open Datasets Yes Our first experiment with real-world data is based on the UCI Student Performance Data Set (Cortez & Silva, 2008) and The task of credit risk assessment involves predicting the likelihood of a borrower defaulting on a loan. For our experiment, we utilize the Credit Risk Dataset... The causal graphs for the Credit Risk Dataset is provided in Figure 13. The attribute information can be found at https://www.kaggle.com/datasets/laotse/credit-risk-dataset.
Dataset Splits Yes The proportion of training, validation and test data is split as 8 1 1.
Hardware Specification No The paper mentions using 'the LIEF HPC-GPGPU Facility hosted at the University of Melbourne' but does not specify exact hardware components like GPU models, CPU models, or memory details.
Software Dependencies No The paper mentions software like TETRAD and GES algorithm, and mixture density networks, but does not specify version numbers for these or other software dependencies.
Experiment Setup Yes The λ in our optimisation problem is [0, 0.5, 5, 20, 60, 100]. and Additionally, in our ϵ-IFair model, the λ is set to be [0, 1, 40, 100, 130, 175, 250]. and For each model, we run it 20 times with different seeds and report the average results in the main section.