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].

When Worlds Collide: Integrating Different Counterfactual Assumptions in Fairness

Authors: Chris Russell, Matt J. Kusner, Joshua Loftus, Ricardo Silva

NeurIPS 2017 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the flexibility of our method on two real-world fair classification problems: 1. fair predictions of student performance in law schools; and 2. predicting whether criminals will re-offend upon being released. For each dataset we begin by giving details of the fair prediction problem. We then introduce multiple causal models that each possibly describe how unfairness plays a role in the data. Finally, we give results of Multi-World Fairness (MWF) and show how it changes for different settings of the fairness parameters (ϵ, δ). We split the law school data into a random 80/20 train/test split and we fit casual models and classifiers on the training set and evaluate performance on the test set.
Researcher Affiliation Academia Chris Russell The Alan Turing Institute and University of Surrey EMAIL Matt J. Kusner The Alan Turing Institute and University of Warwick EMAIL Joshua R. Loftus New York University EMAIL Ricardo Silva The Alan Turing Institute and University College London EMAIL
Pseudocode Yes Algorithm 1 Multi-World Fairness
Open Source Code No The paper does not provide any explicit statement or link regarding the availability of its source code.
Open Datasets Yes We begin by investigating a dataset of survey results across 163 U.S. law schools conducted by the Law School Admission Council [19] and Pro Publica [13] released data on prisoners in Broward County, Florida who were awaiting a sentencing hearing.
Dataset Splits No The paper mentions an “80/20 train/test split” for both datasets but does not specify a validation split.
Hardware Specification No The paper does not provide any specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies, such as programming language versions or library version numbers (e.g., Python, TensorFlow, PyTorch versions).
Experiment Setup No The paper mentions using linear classifiers and a 3-layer neural network and describes how λ values are selected through grid search. However, it does not provide specific hyperparameter values like learning rates, batch sizes, number of epochs, or optimizer settings for the models used in the experiments.