CS 740 - Mathematical Methods for Visual Computing



Course Information

Course Description

This course will cover basic concepts of mathematical methods that are relevant to visual computing (Graphics, Computer Vision, Image Processing) as well as AI/ML. The topics covered include basic numerical linear algebra, probability and statistics, and optimization. This course is at a preparatory level to get first year graduate students (MS, MTech, PhD) introduced to these topics, so that it helps them in more advanced courses. This course is not intended to go very deep into any of these areas, but is intended to familiarize students with the material. A tentative list of topics we will cover is as follows. Alongside each topic, relevant applications in machine learning/computer vision/image processing/computer graphics will be mentioned.

Intended Audience and Pre-requisites

First year Mtech, MS, PhD students from CSE, EE, IEOR, Syscon, Mathematics, KCDH, CMiNDS. This course is NOT for UG students. If you are a UG student, please do NOT register.

Textbooks and Learning Material

Resources


Grading scheme (tentative)

Course project

Project Topics and Instructions

Detailed Schedule

Date

Content of the Lecture

Assignments/Readings/Notes

Lecture 1: 27/7 (Mon)
  • Course overview
Lecture 2: 28/7 (Tue)
    Numerical Linear Algebra
  • Matrix: trace, determinant, rank, matrix multiplication
  • Geometric meaning of a matrix, systems of linear equations and a linear least squares problem
Lecture 3: 30/7 (Thurs)
  • Translations as a matrix
  • Derivation of linear least squares
Lecture 4: 1/8 (Sat)
  • Derivation of linear least squares
  • Linearly dependent and linearly independent vectors, rank of a matrix
Lecture 5: 3/8 (Mon)
  • MATLAB tutorial: code vectorization, matrix and vector operations, plotting of functions (plot, surf, surfc)
Lecture 6: 4/8 (Tue)
  • Matrix rank
  • Nullspace and column-space of a matrix; homogeneous linear systems
  • Orthogonal matrices
Lecture 7: 6/8 (Thurs)
  • Orthogonal matrices
  • Discrete Cosine transform matrix
  • Vector norms
Lecture 8: 8/8 (Sat)
  • Vector norms
  • Matrix norms, induced norms, condition number, stability of a linear system Ax = b given errors in b
Lecture 9: 10/8 (Mon)
  • Stability of a linear system Ax = b given errors in A
  • Eigenvectors and eigenvalues, eigenvvalue multiplicity, physical examples of eigenvectors and eigenvalues
Lecture 10: 11/8 (Tue)
  • Properties of eigenvectors and eigenvalues
  • Power iteration algorithm for finding the dominant eigenvector
Lecture 11: 13/8 (Thurs)
  • Properties of eigenvectors and eigenvalues of symmetric matrices
  • Singular value decomposition of a matrix: introduction
Lecture 12: 24/8 (Mon)
  • Singular value decomposition of a matrix: introduction, derivation
  • Eckart-Young theorem, SVD in image compression, reduced form of SVD, geometric interpretation of SVD
Lecture 13: 25/8 (Tue)
  • Orthogonal procrustes algorithm: derivation, extension to estimate translations (with derivation); extension for the case of rotation matrices (result without derivation)
Lecture 14: 27/8 (Thurs)
  • SVD applications for: pseudo-inverse or inverse, rank, matrix nullspace, Frobenius norm
  • Introduction to structure from motion
Lecture 15: 31/8 (Mon)
  • SVD application for nullspace computation
  • Introduction to structure from motion: rank theorem, use of SVD
Lecture 16: 1/9 (Tue)
  • Structure from motion: rank theorem, use of SVD, algorithm
  • LU decomposition/Gaussian elimination for solving linear simultaneous equations
Lecture 17: 3/9 (Thurs)
  • Cholesky decomposition, solving banded linear systems
  • Principal Components Analysis: concept, derivation, applications to face recognition
Lecture 18: 7/9 (Mon)
  • Principal Components Analysis: applications to face recognition, efficient algorithm that side-steps creation of a covariance matrix, choice of k using cross-validation, derivation of multiple, mutually perpendicular directions
  • Concept of eigenvalue of covariance matrix as the variance of an eigencoefficient
Lecture 19: 8/9 (Tue)
  • Principal Components Analysis: applications to face recognition
  • QR decomposition of a matrix and Gram-Schmidt orthogonalization
  • Cholesky decomposition of a positive semi-definite matrix