Kernekoncepter
Conformal Predictive Systems (CPS) can be extended to handle covariate shifts between training and test data by leveraging likelihood ratios between their covariate distributions.
Resumé
The paper introduces Weighted Conformal Predictive Systems (WCPS) and Weighted Split Conformal Predictive Systems (WSCPS) to address covariate shifts in predictive modeling.
Key highlights:
- CPS offer a versatile framework for constructing calibrated predictive distributions, but their applicability has been limited to scenarios adhering to the Independent and Identically Distributed (IID) assumption.
- The authors propose WCPS and WSCPS, which extend CPS to accommodate scenarios characterized by covariate shifts by leveraging likelihood ratios between training and testing covariate distributions.
- The theoretical underpinnings and conjectures regarding the validity and efficacy of WCPS and WSCPS are presented.
- Simulation experiments on synthetic and real-world datasets indicate that WSCPS are probabilistically calibrated under covariate shift.
- The proposed methods offer a promising avenue for addressing covariate shifts in predictive modeling, with potential applications in diverse fields.
Statistik
The likelihood ratio between training and test covariate distributions is given as w(x) = exp(-x1 + 0.5x2 - 0.25x3 - 0.1x4) for the synthetic data.
The effective sample size of the calibration set for WSCPS is computed as ˆn = [Σni=1 |w(xi)|]2 / Σni=1 |w(xi)|2.
Citater
"Conformal Predictive Systems (CPS) offer a versatile framework for constructing predictive distributions, allowing for calibrated inference and informative decision-making."
"We therefore propose Weighted CPS (WCPS), akin to Weighted Conformal Prediction (WCP), leveraging likelihood ratios between training and testing covariate distributions."
"Our simulation experiments indicate that WCPS are probabilistically calibrated under covariate shift."