Tutorial on Graphical Models
![]()
An ICCV 2015 Tutorial onInference and Learning in Discrete Graphical Models:
|
8:30 | Opening |
8:40 | Discrete Graphical Models
|
9:30 | Inference in Discrete Graphical Models : Part 1
|
10:00 | Coffe break |
10:30 | Inference in Discrete Graphical Models : Part 2
|
12:00 | Lunch |
14:00 | From Benchmarks to the Current Limits
|
15:00 | Coffe break |
15:30 | Learning in Discrete Graphical Models
|
16:20 | Closing |
Organizers |
|
![]() Bogdan Savchynskyy Computer Vision Lab Dresden, Technische Universität Dresden, Germany |
graduated from the National Technical University of Ukraine ”Kiev Polytechnic Institute” in 2002. In 2007 he defended |
![]() Jörg Hendrik Kappes Heidelberg Collaboratory for Image Processing, Heidelberg University, Germany |
received diploma degree in computer science from the University of Mannheim in 2005 and it PhD degree from the Heidelberg University 2011. |
![]() Thorsten Beier Multidimensional Image Processing Group, Heidelberg University, Germany |
received bachelor degree in physics from the University of Heidelberg in 2011, |
![]() Sebastian Nowozin Microsoft Research Cambridge, UK |
is researcher in the Machine Learning and Perception (MLP) group at Microsoft Research, Cambridge, UK. |
![]() Carsten Rother Computer Vision Lab Dresden, Technische Universität Dresden, Germany |
received the diploma degree in 1999 from the University of Karlsruhe, Germany, and the PhD degree in 2003 from |
Tutorial DescriptionThe proposed tutorial will focus on inference in discrete graphical models and its application in computer vision. In particular, we will cover globally optimal methods provided by general ILP-solvers, which recently have become For linear programming relaxations we will give the fundamental theory and show relations between the corresponding dedicated solvers. The part of the tutorial entitled Approximative and Move Making Methods will be devoted to such methods as alpha-expansion, We will conclude the inference part of the tutorial by showing how different methods can be combined into meta-methods, which often have the best overall performance. An important part of this tutorial are common use-cases, problems and pitfalls, which will be covered by a separate lecture block with several live demos. We then will turn to the problem of learning parameters of discrete graphical models. |
Downloads
inference.pdf (32Mb)
learning.pdf (5Mb)