Statisztikus tanuláselmélet: konformális predikció
| Témavezető: | Csáji Balázs Csanád |
| SZTAKI és ELTE TTK Valószínűségelméleti és Statisztika Tanszék | |
| email: | csaji@sztaki.hu |
Projekt leírás
Predicting the outputs of an unknown system from noisy measurements is a fundamental problem in statistics and machine learning (Shalev-Shwartz and Ben-David, 2014). Conformal prediction (Vovk, Gammerman, and Shafer, 2005) is a relatively new machine learning framework designed to quantify the uncertainty of almost any underlying prediction model. It fundamentally achieves this by employing resampling techniques to convert an algorithm’s outcomes into prediction sets with strong, finite-sample coverage guarantees (Tibshirani, 2023). Key advantages of conformal prediction include that its guarantees are distribution-free (besides non-asymptotic) and it can be combined with a vast array of statistical and machine learning approaches. The aim of this project is to investigate, both theoretically and experimentally, the effectiveness of conformal prediction in specific problem classes with characteristic structures, starting with simple linear regression.
Hivatkozások
- Ryan Tibshirani: Conformal Prediction, "Advanced Topics in Statistical Learning" Course, University of California, Berkeley, Spring 2023.
- Vladimir Vovk, Alex Gammerman, and Glenn Shafer: "Algorithmic Learning in a Random World", Springer, 2005.
- Shai Shalev-Shwartz, and Shai Ben-David: "Understanding Machine Learning: From Theory to Algorithms", Cambridge University Press, 2014.