Uncertainty Quantification

Sangdon Park
Sangdon Park
ChatGPT 4o

Uncertainty quantification is a critical tool for building trustworthy AI. We consider the following key question:

Can we rigorously learn and quantify the uncertainty of AI models, e.g., Large Language Models (LLMs), price predictors, or drones, under distribution shift and adversarial manipulation?

On-going/Potential Projects

  • Selective Prediction: Improve selective prediction methods.
  • Conformal Prediction: Improve conformal prediction methods.
  • Calibration: Improve calibration methods.

Keywords

  • Uncertainty Quantification
  • Calibration
  • Conformal Prediction
  • Learning Theory
  • Distribution Shift