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].
Regression under demographic parity constraints via unlabeled post-processing
Authors: Gayane Taturyan, Evgenii Chzhen, Mohamed Hebiri
NeurIPS 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our algorithm is backed by finite-sample analysis and post-processing bounds, with experimental results validating our theoretical findings. |
| Researcher Affiliation | Collaboration | Gayane Taturyan IRT System X, Université Gustave Eiffel, Université Paul-Sabatier EMAIL Evgenii Chzhen CNRS, Université Paris-Saclay EMAIL Mohamed Hebiri Université Gustave Eiffel EMAIL |
| Pseudocode | Yes | Algorithm 1: DP post-processing(L, T, β, p, B, η, τ) |
| Open Source Code | Yes | The code is available at https://github.com/taturyan/unaware-fair-reg. |
| Open Datasets | Yes | We conduct our study on two datasets: Law School dataset (Wightman (1998)) and Communities and Crime dataset (Redmond (2009)). In both datasets, ethnicity is a sensitive attribute [...] Additional experiments on Adult dataset (Lichman (2013)). |
| Dataset Splits | Yes | First, we randomly split the data into training, unlabeled and testing sets with proportions of 0.4 0.4 0.2. |
| Hardware Specification | Yes | The experiments are conducted on a Processor 11th Gen Intel(R) Core(TM) i7-1195G7 2.90GHz with 16GB RAM. |
| Software Dependencies | No | We use simple Linear Regression and Logistic Regression from scikit-learn for training the regressor and the classifier. While 'scikit-learn' is mentioned, no specific version number is provided for it or any other key software component. |
| Experiment Setup | Yes | We illustrate the convergence for ε = (2 8, 2 8) unfairness threshold. [...] We train Communities and Crime dataset for N=15000 iterations and Law School dataset for N=5000 iterations for each pair of epsilons. We use parameters L = T and B = 1 for both datasets. |