ELEC 446: Computational Inference

ELEC 446 develops Computational Inference: Bayesian inference for uncertainty quantification, stochastic modelling and MCMC sampling, advanced Monte Carlo strategies, imaging and machine vision.

Semester: 2

Recommended Preparation: Inverse Problems and Imaging module

Lecturer: Colin Fox. Room 503. Ph 4797806. colin.fox 'at' otago.ac.nz

Assessment: 30% assignments, 70% Project (ELEC 446 will be internally assessed in 2016)

Lectures and Tutorials: There is a 2-hour lecture and 1-hour tutorial per week

Resources:

• Lecture 1: Problem and Goals
  • ELEC 445 Inverse Problems course notes
  • Gareth Roberts' notes on Statistical Inference
• Lecture 2: Evaluating Expectations, MH MCMC proposals in few dimensions
• Lecture 3: Adaptive MCMCs and Efficiency
  • IA2RMS: Martino Read Luengo 2015 paper and Matlab package
  • AM: Roberts and Rosenthal 2009 paper and Marko Laine's MCMC toolbox
  • Ulli Wollf's UWerr MatLab code and documentation
• Lecture 4: Inverse Diffusion Problem
• Lecture 5: Image Models
• Lecture 6: Linear-Gaussian Inverse Problems
  • Paper on MTC for image deblurring
• Lecture 7: Approximations
  • Paper on MCMC using an approximation, ADAMH
• Lecture 8: Counting objects
• Assignment 1:
  • Problem sheet
  • Matlab code for diffusion problem: Main.m heat.m
  • Useful code snippets: tauvsd.m (an example of running AM on multivariate normals), stdnorm_example.m (an example of using IA2RMS to sample from N(0,1)), mcgaus.m (RWM for N(0,1)).
• Assignment 2:
  • Problem sheet
  • Matlab code for deblurring jupiter image: jupiter1.tif jupiter.m deconv1.m
• Assignment 3:
  • Problem sheet
  • Image of good and bad cells in Matlab mat format slide.mat, or as slide.tif. Matlab code makefake.m that generated the image, and uses functions putgood.m and putbad.m.
• Project:

Links:

Module 412 in 2015, BUC5 compute resources, Blockkurs on Markov Chain Monte Carlo for Inverse Problems in PDEs

Department of Physics 400-level information