Fairness in Forecasting of Observations of Linear Dynamical Systems
Authors: Quan Zhou, Jakub Mareček, Robert Shorten
JAIR 2023 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our empirical results on a biased data set motivated by insurance applications and the well-known COMPAS data set demonstrate the efficacy of our methods. |
| Researcher Affiliation | Academia | Quan Zhou EMAIL Dyson School of Design Engineering, Imperial College London, London, SW7 9EG, United Kingdom School of Electrical and Electronic Engineering, University College Dublin, Dublin, D04 V1W8, Ireland Jakub Mareček EMAIL Department of Computer Science, Czech Technical University in Prague, Prague, 121 35, the Czech Republic Robert Shorten EMAIL Dyson School of Design Engineering, Imperial College London, London, SW7 9EG, United Kingdom School of Electrical and Electronic Engineering, University College Dublin, Dublin, D04 V1W8, Ireland |
| Pseudocode | No | The paper describes mathematical formulations (Equations 2-7) and computational methods (TSSOS, NPA hierarchy) but does not include any clearly labeled pseudocode or algorithm blocks for the proposed methods. |
| Open Source Code | Yes | Our implementation is available on-line at https://github.com/Quan-Zhou/Fairness-in-Learning-of-LDS. |
| Open Datasets | Yes | We tested our formulations on the famous Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) dataset. ... Downloaded from https://github.com/propublica/compas-analysis, this dataset is what (Angwin et al., 2016) used in analysing the racial bias in COMPAS recidivism scores. |
| Dataset Splits | Yes | For a single trial, we randomly pick 80% of samples as the training set then test the output on the rest 20% of the samples. |
| Hardware Specification | Yes | implemented with five CPUs and 64GB of memory per CPU. |
| Software Dependencies | Yes | The SDP of a given order in the respective hierarchy can be constructed using ncpol2sdpa 1.12.22 of (Wittek, 2015) or the tools of (Wang & Magron, 2021) 3 and then solved by mosek 9.2 of (MOSEK, Ap S, 2020). |
| Experiment Setup | Yes | The three models in Equations (5)-(7) are applied in each experiment with λ1 of 1, 3, and 5, respectively, as chosen by iterating over integers 1 to 10, while λ2 remains 0.01, The mean of forecast ft across 10 experiments and its standard deviation are shown as solid curves with error bands. |