	Log in \| Register \|

NIPS '07 Workshop on Approximate Bayesian Inference in Continuous/Hybrid Models

Infer.NET - Practical Implementation Issues and a Comparison of Approximation Techniques

author: John Winn, Microsoft Research, Cambridge

Description

Infer.NET is an efficient, general-purpose inference engine developed at Microsoft Cambridge by Tom Minka, John Winn and others. It aims to be highly efficient, general purpose and extensible --- three normally contradictory goals. We have largely managed to achieve these goals using a compiler-like architecture, so that code is generated to perform the desired inference task. Infer.NET can apply one of a range of inference algorithms to a given probabilistic model, and so provides a useful framework for comparing the performance of different algorithms. In this talk, I will describe the capabilities and infrastructure of Infer.NET and give examples of applying both expectation propagation and variational message passing on the same model. I will also describe some failure cases that we have encountered for each algorithm.

You might be experiencing some problems with Your Video player.

*Slides*
0:00	Expectation Propagation & Variational Message Passing
0:55	- Deterministic approximate inference
1:48	Divergence minimisation - 1
4:03	- Questions
7:10	Message passing
10:07	- Comparing EP and VM
10:22	Beta‐Bernoulli
13:25	Gaussian with fixed variance
14:51	Surprising conclusion
15:33	- Questions
21:29	Learning mean and precision - 1
22:07	Learning mean and precision - 2
22:35	Learning mean and precision - 3
23:10	Learning mean and precision - 4
24:30	So...
24:56	- Questions
27:11	Learning mean and precision - 4
27:30	Improper messages are necessary
28:29	- Questions
37:27	Product of Gaussians - 1
38:22	Product of Gaussians - 2
39:47	Product of Gaussians - 3
40:46	Exact posterior for A
41:08	Posterior for A vs. variance of B
42:14	- Questions
42:22	Posterior for A vs. variance of B
43:42	- Questions
45:14	Demo #3
45:25	Summary
45:34	- Questions

Lecture rating

People found this lecture:
Worth seeing
because it is:
Valuable and informative
Well presented
Easily understandable
Acceptably recorded
You need to login to cast your vote.

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.