CS 504 Computational Modeling


Computational models are executable explanations of complex phenomena. In this course, we will build and debug computational models that implement theories of perception, action, behavior and visual appearance. We will explore computational techniques that use models to guess (estimate) the current state of the world given observations of it. And we will review methods to assess and evaluate these models.

Class Meetings

Thursday 1/24 - Introduction

Thursday 1/31 - An overview of models

Thursday 2/7: Continuous simulation

Thursday 2/14: Discrete simulation

Thursday 2/21 - Simulation of light

Thursday 2/28 - Discrete models and inference

Thursday 3/7 - Continuous models and inference

Thursday 3/14

Thursday 3/21

Thursday 3/28 - Overview: Sampling for probabilistic inference

  • external link: Slides
  • Classic paper on particle filters for estimation (Condensation): external link: PDF Dellaert, Burgard, Fox and Thrun, Using the CONDENSATION algorithm for robust, vision-based mobile robot localization. CVPR 1999.
  • Another useful application of particle filters external link: PDF De Freitas, Rao-Blackwellised particle filtering for fault diagnosis. IEEE Aerospace Conference, 2002.
  • external link: Tarball of de Freitas's matlab demo code for particle filtering. Originally external link: here.

Thursday 4/4 - Understanding Sampling - Continuous domains

Thursday 4/11 - Understanding Sampling - Discrete domains

Thursday 4/18 - Mixing discrete and continuous: Clustering

  • Processing image data: external link: PCA
  • Expectation maximization clustering: mixtures of Gaussians: external link: tutorial reading and external link: Matlab code from external link: here. (Tutorial is Theory and Use of the EM Algorithm, by Maya R. Gupta and Yihua Chen, Foundations and Trends in Signal Processing Vol. 4, No. 3 (2010) 223–296.)
  • external link: Matlab code to cluster image pixels with a mixture of Gaussian model and Expectation Maximization.
  • Bayesian Clustering with the Chinese Restaurant Process. external link: application paper and external link: Matlab code from external link: Frank Wood. Paper is Wood and Black, A Nonparametric Bayesian Alternative to Spike Sorting, Journal of Neuroscience Methods, 173:1-12, 2008.
  • external link: Matlab code to cluster image pixels with a Dirichlet Process Mixture Model and Gibbs Sampling.

Thursday 4/25 - Applications in Vision, I

Thursday 5/2 - Applications in Vision, II