Google, IBM & Meta Certificates – 40% Off
One plan covers every Professional Certificate on Coursera.
Unlock All Certificates
Explore the theoretical foundations of machine learning applications in discrete optimization through this conference talk that examines learning-augmented algorithms combining worst-case guarantees with machine-learned prediction efficiency. Discover how calibration—the alignment between predicted probabilities and observed frequencies—serves as a principled approach to determine optimal trust levels in predictions, moving beyond single global trust parameters to leverage instance-specific uncertainty estimates from modern predictors. Learn through two detailed case studies: ski rental problems and online job scheduling, where the speaker demonstrates how calibrated advice achieves near-optimal prediction-dependent performance and outperforms alternative uncertainty quantification methods, particularly in high-variance settings. Gain insights into experimental validation using real-world datasets that confirm theoretical results and showcase the practical value of calibration in learning-augmented algorithm design, based on collaborative research with Judy Hanwen Shen and Anders Wikum.
