Recorded Lectures

The lectures are now available on YouTube.

  • Lecture 1: Introduction to Probabilistic Modelling and Machine Learning [1][2][3]
  • Lecture 2: Graphical models: inference and structure learning [1][2]
  • Lecture 3: Gaussian processes and Bayesian kernel machines [1][2]
  • Lecture 4,5: Bayesian non-parametrics [1][2]

Visiting Professor

We are happy to announce that Professor Zoubin Ghahramani (University of Cambridge) is a Visiting Professor at Wroclaw University of Technology from 24.06.2013 to 28.06.2013. He is going to give a series of lectures about machine learning and nonparametric Bayesian statistics. The lectures are public and all interested are welcome. The detailed schedule is given below.

Z przyjemnością informujemy, że Profesor Zoubin Ghahramani (University of Cambridge) będzie Profesorem Wizytującym na Politechnice Wrocławskiej od 24.06.2013 do 28.06.2013. Wygłosi serię wykładów nt. uczenia maszynowego i nieparametrycznej statystyki bayesowskiej. Wykłady mają charakter otwarty i zapraszamy wszystkich zainteresowanych. Szczegółowy harmonogram załączono poniżej.

Lectures Schedule


Subject Time/Place Details Slides
1 Introduction to Probabilistic Modelling and Machine Learning MON
(24.06.2013)

TIME
15.00-18.00

ROOM
C-3, s. 22
I will introduce the probabilistic framework for modelling. Probability theory provides a foundation for uncertainty in artificial intelligence, for learning from data, and for modelling complex systems. The framework is based on a few simple rules which I will review. Problems of inference from data can be solved via the application of inverse probability in the form of Bayes Rule. Topics I will cover include: foundations, choice of priors, model selection, and approximate inference, including MCMC, variational methods and Expectation Propagation. [1]
2 Graphical models: inference and structure learning TUE
(25.06.2013)

TIME
15.00-18.00

ROOM
C-3, s. 22
Probabilistic graphical models are a way of representing independence structure in distributions over many variables. I will review undirected, directed and factor graph frameworks for representing conditional independence, and message passing algorithms for inference. I will then talk about how we can learn the structure of a graphical model from data, and touch upon advanced topics such as causality and structure learning in undirected models. [2a]

[2b]
3 Gaussian processes and Bayesian kernel machines WED
(26.06.2013)

TIME
8.30-11.30

ROOM
C-3, s. 22
Kernel machines such as support vector machines (SVMs) have been extremely successful in machine learning. I will review Gaussian processes (GPs), which are a probabilistic discriminative model directly analogous to SVMs. GPs can be used for regression, classification, ordinal regression, ranking, and many other problems requiring inference on unknown functions. I will discuss learning GPs, including how to learn the kernel from data, and how to scale up learning to large data sets. [3a]

[3b]
4 Bayesian non-parametrics I: Foundations, Dirichlet process, Chinese restaurant process, infinite HMMs, and Indian Buffet process THU
(27.06.2013)

TIME
15.00-18.00

ROOM
C-3, s. 22
Non-parametrics make it possible to have flexible models that are adequate for modelling complex real-world phenomena. Gaussian processes and Dirichlet processes (DPs) are two of the cornerstones of non-parametrics. I will review DPs and their relation to Chinese restaurant processes (CRPs) for clustering and density estimation. I will then describe the extension to time series in the form of infinite Hidden Markov models, and models for overlapping clusters and sparse binary matrices based on the Indian buffet process (IBP). [4]
5 Bayesian non-parametrics II: Models for networks, relational data, hierarchies, sparsity, covariances, and deep networks FRI
(28.06.2013)

TIME
8.30-11.30

ROOM
C-3, s. 22
Using the tools from lectures 3 and 4, I will describe how one can develop nonparametric models for social and biological networks and other relational data, models for hierarchical clustering, models for sparse factor analysis and matrix factorisation, and models for spatially evolving covariance matrices. I will conclude by showing how we can learn the structure of deep layered belief networks from data using Bayesian non-parametrics. [5]

Contact Details

E-MAIL

adam.gonczarek@pwr.edu.pl

OFFICE ROOM

Building: C-3

Room: 121

CORRESPONDENCE ADDRESS

Department of Computer Science,
Wroclaw University of
Science and Technology
27 Wybrzeże Wyspiańskiego St
50-370 Wrocław, Poland

TELEPHONE

+48 71 320 4453