University of Cambridge > Talks.cam > Machine Learning @ CUED > Developments in Exact Inference in Graphical Models

Log in

University Account

External (via Google)

Information on

Subscribing to talks details

Finding a talk details

Adding a talk details

Disseminating talks details

Help and Documentation details

Developments in Exact Inference in Graphical Models

Download to your calendar using vCal

Stephen Pasteris (UCL)
Tuesday 07 June 2016, 10:30-11:30
Engineering Department, CBL Room BE-438.

If you have a question about this talk, please contact Zoubin Ghahramani .

My talk is split into two parts:

Part 1 - I develop an algorithm for dynamic updating and marginalisation in tree-structured Markov Random Fields. This algorithm is always at least as fast as any other algorithm and in some cases is exponentially faster than any other algorithm. The initialisation time required by the algorithm is the optimal, linear time.

Part 2 - The junction tree algorithm can be slow when we have large supernodes of high degree and when we have such supernodes the difference in time complexity between different architectures becomes very obvious. Previously, the fastest junction tree algorithms had a very large memory requirement. I develop a junction tree architecture which has, essentially, state of the art speed (faster than HUGIN propagation) while essentially maintaining the low space requirement of the slower Shafer-Shennoy propagation.

This talk is part of the Machine Learning @ CUED series.

This talk is included in these lists:

Note that ex-directory lists are not shown.

Other lists

Gurdon Institute Tearoom MR13, CMS Pavilion E Webinar & LT2, Computer Laboratory, William Gates Building.

Other talks

Toward high resolution structure/function analyses of sensorimotor integration in ciliated protists The concept of fitness in evolutionary biology: a philosophical analysis Afternoon Break Title TBC Nationalising Oil and Knowledge in Iran: Labour, Decolonisation and Colonial Modernity, 1933–51 Grand Rounds - Equine

Abstract

My talk is split into two parts:

Part 1 - I develop an algorithm for dynamic updating and marginalisation in tree-structured Markov Random Fields. This algorithm is always at least as fast as any other algorithm and in some cases is exponentially faster than any other algorithm. The initialisation time required by the algorithm is the optimal, linear time.

Part 2 - The junction tree algorithm can be slow when we have large supernodes of high degree and when we have such supernodes the difference in time complexity between different architectures becomes very obvious. Previously, the fastest junction tree algorithms had a very large memory requirement. I develop a junction tree architecture which has, essentially, state of the art speed (faster than HUGIN propagation) while essentially maintaining the low space requirement of the slower Shafer-Shennoy propagation.

Log in

🔐 Log In

Information on

ℹ️ Information

Developments in Exact Inference in Graphical Models

This talk is included in these lists:

Developments in Exact Inference in Graphical Models

Abstract

Included in Lists