Computational Science lectures are open to the public and are held at the Gerald D. Fischbach Auditorium at the Simons Foundation headquarters in New York City. Tea is served prior to each lecture.
In this lecture, Mauro Maggioni will discuss a family of ideas, algorithms and results for learning from high-dimensional data.
Andrew Gelman will illustrate this concept with various examples from his recent research and discuss more generally how statistical methods can help or hinder the scientific process.
Probabilistic topic models provide a suite of tools for analyzing large collections of electronic documents. A traditional topic model analyzes a collection of documents to discover its hidden themes. These themes can be used to organize, visualize, summarize and navigate the collection. Many collections are associated with corresponding reader behavior data, which is useful both for making predictions about readers (such as which articles they will like) and in understanding patterns in how they read.
Learning Using Privileged Information (LUPI) is a new paradigm that uses an intelligent agent (a ‘nontrivial teacher’) to supplement standard training data in the context of supervised learning algorithms. Rather than using standard, brute-force methods to address the general problem of inference and the construction of intelligent machines, the LUPI learning model allows the teacher to add additional (privileged) information to the training examples.