Probabilistic Inference Group (PIGS) - Archive

Meetings in 2010

Tue 14 December (Peter Orchard)

We will discuss the following paper:

Tue 30 November (Iain Murray)

We will discuss the following paper:

Tue 16 November (Amos Storkey)

Amos will give an overview of a number of papers around the subject of sequential Monte-Carlo, particle MCMC and inference and parameter estimation for continuous time systems.

This session will be more of a presentation than a paper reading and is intended to give a basic encapsulation of a number of papers, thus covering several topics in summary.

Tue 2 November (Andrew Dai)

We will discuss the following papers:

Tue 19 October (Guido Sanguinetti)

We will discuss the following paper:

Tue 5 October (Jyri Kivinen)

We will discuss the following papers:

Tue 21 September (Ali Eslami)

We will discuss the following papers:

Tue 7 September (Athina Spiliopoulou)

We will discuss the following paper:

Tue 24 August (Frank Dondelinger)

We will discuss the following paper:

Tue 10 August - ICML / UAI / ISBA 2010

Tue 27 July - ICML / UAI / ISBA 2010

Tue 13 July (Jakub Piatkowski)

We will discuss the following paper

Tue 29 June (Chris Williams)

AISTATS 2010 feedback:

Plus quick reviews for:

Tue 18 May (Charles Sutton)

We will discuss the following paper

Tue 4 May (Peter Orchard)

We will discuss the following paper

Tue 20 April - Kathrine Heller

Talk by Kathrine Heller. Please see title and abstract below.

Beyond basic clustering: Bayesian statistical methods for heterogeneous data

In this talk I will focus on using Bayesian methods to model data in situations where we would like to discover latent cluster structure, but where our data is too heterogeneous to be properly modeled using standard clustering algorithms, which assume that data belong exclusively to a single cluster.

In the first half of this talk, I'll discuss the IBP Compound Dirichlet (ICD) process, a nonparametric Bayesian method which is an alternative to the Hierarchical Dirichlet Process (HDP) for mixed-membership modeling. In this work, we address the HDP assumption that the probability of a mixture component (ie cluster) contributing to a data point is positively correlated with the amount that it contributes. In many settings, this is an undesirable prior assumption. For example, in topic modelling a topic (component) might be rare throughout the corpus but dominant within those documents (data points) where it occurs. The ICD integrates features from both the HDP and the Indian buffet process (IBP).We use an ICD mixture model, the focused topic model (FTM), to analyze text corpora, and demonstrate superior performance over the HDP-based topic model.

In the second half of this talk, I'll discuss Bayesian Sets, a Bayesian method for performing information retrieval, based on queries formed from examples. Here retrieved examples can be seen as belonging to the same cluster as the examples from the query set. This retrieval process performs "clustering on demand", where a potentially unbounded
number of alternative clusterings of the data exist (through different queries). We formulate this as a Bayesian inference problem and describe a very simple and efficient algorithm for solving it. We use this Bayesian information retrieval system to perform content-based image retrieval, as well as demonstrations on multiple other types of data.

Tue 6 April (David Reichert)

We will discuss the following papers

Tue 23 March (Amos Storkey)

We will discuss the following paper

Tue 23 February (Andrew Dai)

We will discuss the following papers:

Tue 9 February - David Barber

Talk by David Barber. Please see title and abstract below:

Title: New applications for some old variational tricks

Abstract: I'll discuss two applications of variational inference. The first regards so-called local variational methods that have been been popular in recent years in Bayesian logistic regression and Bayesian image analysis (MRI). I'll show how these local variational methods are related to weakened global variational methods. The second application is to novel solution methods for Markov Decision Processes, and related issues in reinforcement learning.

Note: This session will be held in IF.4.33

Tue 26 January (2nd NIPS session)

Tue 12 January (NIPS session)

Topic revision: r1 - 12 Jan 2011 - 13:59:50 - AthinaSpiliopoulou
 
This site is powered by the TWiki collaboration platformCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback
This Wiki uses Cookies