CS 886 (Sect.2) - Perception as Bayesian Inference

Winter 2010
Department of Computer Science
University of Waterloo

Instructor: Richard Mann, DC2510, x3006, mannr@uwaterloo.ca

Meetings: Fridays 13:00-16:00, MC2306.
First class:
Friday 8 January

Important Note: Please try to send all course material from a University of Waterloo account (ie., SSH into a teaching or research machine and send from there).  If you don't have a Waterloo account yet, please contact the MFCF help center (MC3011).  In the meantime you can use the account you gave in the first class.  Also please don't send any messages with "zip" or "gif" attachments (send "tif" or "jpg" instead, or refer me to a URL).  Those are common filetypes in Microsoft viruses and are removed by my SPAM filter.


News:



Summary

Advanced topics course.  We will read/evaluate/discuss one or two papers each week.  Students will present one or two papers during the term.  Each student is responsible for preparing a one-page review/commentary of the paper discussed.  Note that I am not looking for a complete summary of the paper.  Instead, I want to know: 1) how does this paper apply Bayesian methods, 2) how does this paper relate to current (or your) research problems.  This commentary should be handed in at the beginning of class where the paper is discussed. The project may be based on currrent research papers, or the student's research area.

Marking (tentative)

References

All required material will be provided in lectures.  We will also draw on the following book (edited collection of book chapters): Other books:

Lecture Schedule (preliminary)

Lecture
Topic
References
Homework
1. (Jan 8)

Organizational meeting.  Overview. Overview paper: Object Perception as Bayesian Inference,
Kersten et al, Ann. Rev. Psychol. 2004, 55:271-304.
Read paper to prepare for course.
2. (Jan 15)
Image alignment by maximum likelihood.
Reference: Registering a Multi-sensor Ensemble of Images,  Orchard and Mann.  To appear, IEEE Transactions on Image Processing, 2010.
Supplementary material: Alignment by Maximization of Mutual Information.  IJCV 24(2):137-154 (1997).

Read paper and write one page summary/commentary (due before class).
3. (Jan 22)
Markov random field models for images.
Reference: Modeling Image Analysis Problems Using Markov Random Fields.  S.Z. Li.  Handbook of Statistics, Vol.20.  Elsevier Science 2000. Pages 1-43.
Additional references: Bayesian Modeling of Uncertainty in Low-Level Vision.  R. Szeliski.  International Journal of Comptuer Vision. 5(3):271-301, 1990.  Stochasstic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images.  IEEE Trans. PAMI. 6(6).  Nov. 1884.  Markov Random Fields with Efficient Approximations.  Y Boykov, O. Veksler, R. Zabih.  CVPR 1998.
Supplementary material:
Read Li and Szeliski.  Write commentary on Li.  Geman and Geman is difficult, but worth looking at.
4. (Jan 29)
Bayesian model comparison, generic viewpoint assumption.
References: Bayesian Interpolation.  David MacKay.  Neural Computation 4, 415-447 (1992).  Exploiting the Generic Viewpoint Assumption.  William T. Freeman.  IJCV 20(3): 243-261 (1996).
Read MacKay first, then Freeman.  Comment on the one of your choice.
5. (Feb 5)
Qualitative probabilities
References: What makes a good feature?.  Allan Jepson and Whitman Richards.  Qualitative probabilites for image interpretatation.  Allan Jepson and Richard Mann.  ICCV (1999).
Read both Jepson/Richards first.  Comment on Jepson/Mann.
6. (Feb 12)
Hidden Markov Models (presenter, Adam Fourney)
Reference: A Tutorial on Hidden Markov models and Selected Applications in Speech Recognition.  Lawrence Rabiner.  Proc. IEEE. 77(2). 1989.
Read and comment on this paper.
 . Reading week.


7 (Feb 26)
Condensation (Darren),
Infinite Mixture model (Ricardo)
Reference #1: CONDENSATION—Conditional Density Propagation for Visual Tracking. M. Isard and A. Blake.  IJCV 29(1):5--28 (1998).
Reference #2: The Infinite Gaussian Mixture Model.  Carl Edward Rasmussen.  NIPS 12 (2000).
Read and comment on both papers.
8 (Mar 5)
CAT scan (Ahmed), Stereo vision (Alexei)
No reference for #1.  Reference #2.Jian Sun; Nan-Ning Zheng; Heung-Yeung Shum; , "Stereo matching using belief propagation," Pattern Analysis and Machine Intelligence, IEEE Transactions on, vol.25, no.7, pp. 787- 800, July 2003
No reading for #1.  Read Stereo paper and comment.
9 (Mar 12)
Optical snow (application of the Fourier transform)
Langer and Mann.  Optical snow.  IJCV 55(1):55-71. 2003.
Read and comment on paper.
10 (Mar 19)
Statistics of Natural Images
Reference #1: D L Ruderman. Origins of Scaling in Natural Images Vision Research 37(23):3385-3398, 1997
Reference #2: B A Olshausen and D J Field.  Natural image statistics and efficient encoding.  Network: Computation in Neural Systems. 7(2):333-339, 1996.
Read and comment on Olshausen.
11 (Mar 26)
Information theory for learning ("infomax")
Reference: A Bell and T Sejnowski.  An Information-Maximization Approach to Blind Separation and Blind Deconvolution. Neural Computation 7:1129-1159, 1995.
Read and comment on paper.
12 (Apr 1).
Note: different time: Thursday 2:30pm,
DC2306 (AI lab).
Student presentations





Additional Reference Material


Resources