The Limitations of Model Retraining in the Face of Performativity
Abstract
We study stochastic optimization in the context of performative shifts, where the data distribution changes in response to the deployed model. We demonstrate that naive retraining can be provably suboptimal even for simple distribution shifts. The issue worsens when models are retrained given a finite number of samples at each retraining step. We show that adding regularization to retraining corrects both of these issues, attaining provably optimal models in the face of distribution shifts. Our work advocates rethinking how machine learning models are retrained in the presence of performative effects.
Cited in this thesis
Frequently Cited Together
- Ordinal regression with multiple output cnn for age estimation1 chapter
- Fishers' preference for mobile traceability platform: challenges in achieving a 1 chapter
- An analytical and machine learning approach for total mercury and methylmercury 1 chapter
- Prediction of coffee traits by artificial neural networks and laser-assisted rap1 chapter
- Multilayer feedforward networks are universal approximators1 chapter
- Support vector learning for ordinal regression1 chapter
BibTeX
@article{Kabra2024,
author = {Kabra, Anmol and Patel, Kumar Kshitij},
journal = {arXiv preprint arXiv:2408.08499},
title = {The Limitations of Model Retraining in the Face of Performativity},
year = {2024},
}