Incentivizing Compliance with Algorithmic Instruments
Authors: Dung Daniel T Ngo, Logan Stapleton, Vasilis Syrgkanis, Steven Wu
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we present experiments to evaluate Algorithm 1. ... In Figure 1, we compare the approximation bound on |θ ˆθ| between IV estimate ˆθ versus via a naive estimate for a specific, chosen ρ = 0.001. ... Results are averaged over 5 runs; light blue error bars represent one standard error. |
| Researcher Affiliation | Collaboration | 1University of Minnesota 2Microsoft Research 3Carnegie Mellon University. |
| Pseudocode | Yes | Algorithm 1 Overcoming complete non-compliance and Algorithm 2 Overcoming partial compliance |
| Open Source Code | Yes | The code is available here. |
| Open Datasets | No | The paper describes generating synthetic data for simulations, stating 'we estimate it using Monte Carlo simulation by running the first stage of Algorithm 1 for 1000 iterations and aggregating the results.' It does not mention using or providing access information for any standard public datasets. |
| Dataset Splits | No | The paper describes numerical simulations and collecting samples, but it does not provide specific details on how data was partitioned into training, validation, or test sets, nor does it specify any cross-validation setups. |
| Hardware Specification | No | The paper includes a 'Numerical Experiments' section but does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used to run these experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9), only referring generally to the code provided. |
| Experiment Setup | Yes | In Figure 1, we let hidden treatment effect θ = 0.5, type 0 and type 1 agents priors on the treatment effect be N(−0.5, 1) and N(0.9, 1) each truncated onto [−1, 1], respectively. We also let the mean baseline reward for type 0 and type 1 agents be µg(0) ∼ N(0, 1) and µg(1) ∼ N(0.1, 1), respectively. |