Speaker
Description
Collider experiments produce vast, high-dimensional datasets where rare signals are buried beneath large and complex backgrounds. To extract reliable conclusions from all available data, it is essential to leverage prior knowledge across domains and to cleverly model the connection between latent variables and observables. In this talk, I will present a set of tools and techniques to build probabilistic models, and show how Bayesian inference can be used to learn latent parameter distributions in an unsupervised way. A fully probabilistic description of data allows for a number of advantages such as well-calibrated soft-assignment probabilities for classification tasks, robust uncertainty quantification, and generative sampling for model validation. Using concrete examples from collider physics, I will illustrate how this approach provides a flexible framework that interfaces modern statistical techniques with fundamental physics problems.
| I read the instructions above | Yes |
|---|