Resumen:
Probabilistic machine learning methods are statistical tools that can be used to make predictions under uncertainty. In this lecture, I will introduce basic concepts of probabilistic machine learning. I will spend the first part of the lecture reviewing the building blocks of probabilistic machine learning along with basic examples, including frequentist and Bayesian viewpoints of probability. I will review Bayes’ theorem and its application to parameter estimation and model comparison. These concepts will be illustrated at the example of simple models such as linear Gaussian systems.
In the second part of the lecture, I will present specific probabilistic models including probabilistic factor analysis models, manifold learning and (generalized) linear regression models.
In this lecture, I will cover the basics concepts underlying the most popular (non-probabilistic) methods for predicting radiation-induced complications, namely support-vector machines and neural networks. In addition, I will give an introduction to feature engineering, feature selection and model selection. I will present approaches, mainly based on cross-validation, that can be used to reliably quantify how well established models would perform on new, unseen data. Pitfalls as well as practical tips will be discussed.
Dirigido a: estudiantes y profesores
Día/Fecha: 28 y 29 de agosto, 2017, 2:00 p.m. – 5:00 p.m.
Localización: Edificio Mac Gregor (N-221)
Profesor responsable e inscripciones: Giancarlo Sal y Rosas (vsalyrosas@pucp.edu.pe)
Nota: La charla será en inglés y la participación es previo registro (cupos limitados).