Improvement-Focused Causal Recourse (ICR)
Authors: Gunnar König, Timo Freiesleben, Moritz Grosse-Wentrup
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In semi-synthetic experiments, we demonstrate that given correct causal knowledge ICR, in contrast to existing approaches, guides toward both acceptance and improvement. and 8 Experiments In the experiments we evaluate the following questions, assuming correct causal knowledge and accurate models of the conditional distributions in the data: Q1: Do CE, CR, and ICR lead to improvement? Q2: Do CE, CR, and ICR lead to acceptance (by pre- and post-recourse predictor)? Q3: Do CE, CR, and ICR lead to acceptance by other predictors with comparable test error? Q4: How costly are CE, CR and ICR recommendations? |
| Researcher Affiliation | Academia | Gunnar König1,2, Timo Freiesleben4,5,6, Moritz Grosse-Wentrup2,3 1 Munich Center for Machine Learning (MCML), LMU Munich 2 Research Group Neuroinformatics, University of Vienna 3 Data Science @ Uni Vienna, Vienna Cog Sci Hub 4 Munich Center for Mathematical Philosophy (MCMP), LMU Munich 5 Cluster of Excellence Machine Learning, University of Tübingen 6 Graduate School of Systemic Neurosciences, LMU Munich |
| Pseudocode | Yes | Details, including pseudocode, are provided in B.1. |
| Open Source Code | Yes | Details on the implementation and access to the code are provided in C.1. |
| Open Datasets | Yes | 7var-covid: A semi-synthetic dataset inspired by a real-world covid screening model (Jehi et al. 2020; Wynants et al. 2020). |
| Dataset Splits | No | The paper mentions 'test error' and 'test set performance' but does not provide specific details on training, validation, or test dataset splits (e.g., percentages or sample counts). It also does not explicitly mention a 'validation' split for data. |
| Hardware Specification | No | The paper does not explicitly describe the hardware (e.g., specific GPU or CPU models, cloud resources) used to run its experiments. |
| Software Dependencies | No | The paper mentions software like 'Random forests', 'logistic regression models', and 'NSGA-II (Deb et al. 2002)' but does not provide specific version numbers for these software components or any other ancillary libraries. |
| Experiment Setup | Yes | For CR and ICR the confidences 0.75, 0.85, 0.9, 0.95 were targeted (for CR: η, for ICR: γ). For CE no slack is allowed, such that the results correspond to a confidence level of 1.0. and For a full specification of the SCMs including the linear cost functions, we refer to C.2. |