Learning under Imitative Strategic Behavior with Unforeseeable Outcomes
Authors: Tian Xie, Zhiqun Zuo, Mohammad Mahdi Khalili, Xueru Zhang
TMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct experiments on both synthetic and real data to validate the theoretical findings (Sec. 5). |
| Researcher Affiliation | Academia | Tian Xie EMAIL Department of Computer Science and Engineering the Ohio State University Zhiqun Zuo EMAIL Department of Computer Science and Engineering the Ohio State University Mohammad Mahdi Khalili EMAIL Department of Computer Science and Engineering the Ohio State University Xueru Zhang EMAIL Department of Computer Science and Engineering the Ohio State University |
| Pseudocode | No | The paper describes methods and theoretical findings but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | code available at https://github.com/osu-srml/Unforeseeable-SC/tree/main |
| Open Datasets | Yes | We conduct experiments on both synthetic Gaussian data and FICO score data (Hardt et al., 2016b). |
| Dataset Splits | No | The paper mentions using FICO data and synthetic Gaussian data, and describes some preprocessing (normalization, estimating distributions) but does not provide specific training/test/validation splits for any dataset. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies or library versions used in the experiments. |
| Experiment Setup | Yes | We use the preprocessed dataset containing the CDF of scores FX|S(x|s), qualification likelihoods PY |XS(1|x, s), and qualification rates αs for four racial groups (Caucasian, African American, Hispanic, Asian). All scores are normalized to [0, 1]. Similar to Zhang et al. (2022), we use these to estimate the conditional feature distributions PX|Y S(x|y, s) using beta distribution Beta(ays, bys). The results are shown in Fig. 9. We assume the improved feature distribution P I(x) Beta a1s+a0s 2 , b1s+b0s 2 and CM CI N(0, 0.25) for all groups, under which Assumption 2.2 and 2.1 are satisfied (see Fig. 8). We also considered other feature/cost distributions and observed similar results. Note that for each group s, the decision-maker finds its own optimal threshold θ s or θ s(ki) or bθ s by maximizing the utility associated with that group, i.e., maxθs E[R(D, Y )|S = s]. |