Stochastic Optimization for Performative Prediction
Authors: Celestine Mendler-Dünner, Juan Perdomo, Tijana Zrnic, Moritz Hardt
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We experimentally explore the trade-off on both synthetic data and a strategic classification simulator. |
| Researcher Affiliation | Academia | University of California, Berkeley {mendler,jcperdomo,tijana.zrnic,hardt}@berkeley.edu |
| Pseudocode | Yes | Figure 1: Stochastic gradient method for performative prediction. Greedy Deploy / Lazy Deploy (contains structured algorithmic steps) |
| Open Source Code | No | The paper does not contain any explicit statement about releasing source code or a link to a code repository. |
| Open Datasets | Yes | Individuals correspond to feature, label pairs (x, y) drawn from a Kaggle credit scoring dataset [9]. ... [9] Kaggle. Give Me Some Credit. https://www.kaggle.com/c/Give Me Some Credit/data |
| Dataset Splits | No | The paper describes how samples are observed and used in the greedy and lazy deploy algorithms, but it does not specify explicit training, validation, or test dataset splits (e.g., percentages or counts) for reproduction purposes. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers (e.g., Python, PyTorch, or specific libraries with their versions). |
| Experiment Setup | Yes | We choose step sizes for both algorithms according to our theorems in Section 3. In the case of lazy deploy, we set = 1, and hence n(k) / k. ... We also experiment with different deployment schedules n(k) for lazy deploy. As described in Theorem 3.3, we can choose n(k) / k for all > 0. ... We choose step sizes for both algorithms as defined in Section 3 with the exception that we ignore the -dependence in the step size schedule of greedy deploy and choose the same initial step size as for lazy deploy (Theorem 3.2). |