The purpose of computing is insight, not numbers. – Richard Hamming

Lecture and section information

CS 4220/CS 5223/MATH 4260, Spring 2022
Lecture time: MWF 11:20-12:10
Lecture location: Olin 255

Staff and office hours

Prof: David Bindel
425 Gates Hall
Phone: 607-255-5395
E-mail: bindel@cornell.edu
OH: TBA (or by appointment)

Catalog description

4 credits. Introduction to the fundamentals of numerical linear algebra: direct and iterative methods for linear systems, eigenvalue problems, singular value decomposition. In the second half of the course, the above are used to build iterative methods for nonlinear systems and for multivariate optimization. Strong emphasis is placed on understanding the advantages, disadvantages, and limits of applicability for all the covered techniques. Computer programming is required to test the theoretical concepts throughout the course.

Prerequisites

Linear algebra at the level of MATH 2210 or 2940 or equivalent and a CS 1 course in any language. We will assume you remember your calculus and can pick up Julia. Recommended but not essential: either CS 3220 or one additional mathematics course numbered 3000 or above. This course can be taken before or after CS 4210/MATH 4250.

Texts

We will cover roughly chapters 1-9.

References

Some linear algebra references:

The Julia language home page has many pointers to using the Julia programming language; see in particular the learning link at the top of the home page.

Approximate lecture schedule

  • Week 1: Matrix manipulations; review of vector spaces, norms, and singular values.
  • Week 2: Sensitivity and conditioning. Floating point.
  • Week 3: Gaussian elimination
  • Week 4: Conditioning, error estimation, and refinement of linear systems. Methods for sparse, banded, and other structured linear systems.
  • Week 5: Introduction to least squares and basic methods.
  • Week 6: Sensitivity, ill-posedness, and regularization in least squares.
  • Week 7: Eigenvalue problem theory, applications, and basic iterations.
  • Week 8: Nonlinear equations and optimization in 1D.
  • Week 9: Stationary iterations and Krylov iterations for linear systems.
  • Week 10: Nonlinear equations, optimization, and multi-dimensional Newton.
  • Week 11: Modified Newton, quasi-Newton, and gradient methods.
  • Week 12: Globalization with line search, trust regions, and continuation.
  • Week 13: Theory and survey of methods for constrained optimization.
  • Week 14: Methods for structured optimization problems. Derivative free methods.
  • Week 15: Review.

Course work

Readings and the problem du jour

Readings from the course text (or notes) will be listed on the course page before class. You are responsible for reading before lecture.

For most lectures, there will be a “problem of the day” related to the class material. These problems are not graded, but you should try them (and try to understand them) as a way to better learn the material – and as a way to study for the exams.

Class work

During each lecture, I will ask 1-2 questions of the class. You should bring along some paper to jot the answers to these questions, as well as notes on anything that you found particularly confusing during the lecture. Each submission counts for a third of a point on the final grade (for a total of 10% of the grade, equal to one project).

It is also critical for us to have your feedback about how the class is going, both to improve the class for the current semester and to make the class better for future semesters. We will solicit non-anonymous comments around the midterm, and at the end of the semester will check with the college to see who has completed course evaluation surveys (though we obviously cannot check to see whether your feedback is useful!). Participating in these feedback activities counts toward your grade via points oon the midterm and final.

Homework and projects

There will be six one-week homeworks, assigned Friday and due the following Friday. These problems will involve a mix of short answers, plots, and computations done in Julia. Homework should be typed and submitted as PDF files on CMS. After they are graded, homework scores will be posted to CMS. Regrade requests must be submitted within one week of receiving the graded homework.

There will be three two-week programming projects, to be done either alone or in a group of two. Projects will involve solving a larger problem, and should be done in Julia. For projects, you will need to submit both codes and a writeup PDF file on CMS.

In order to provide timely, high-quality feedback, we may not always grade all problems in a homework or pieces to a project. Instead, we will focus our grading efforts on providing feedback on a few key points. We will provide written solutions so that you can evaluate yourself for problems where we do not grade in detail.

Paper project

Students taking CS 5223 will be required to read a relevant paper and write a short set of notes on the topic as well as at least two homework-style problems (and their solutions). A proposal for the project should be submitted before the midterm; the project itself may be submitted up to the last day of classes.

Exams

There will be one midterm and one final exam:

  • Midterm: March 6-13
  • Final Exam: May 11-18?

Both exams are take home. A couple points on the final will go toward course evaluation.

Grading

Your final grade in CS 4220 will be computed from grades on the assignments and exams using the following weights:

  • Class work: 10%
  • Homework: 6% times 5 homeworks (best of 6)
  • Projects: 10% times 3 projects
  • Midterm: 15%
  • Final: 15%

If you are taking the course as CS 5223, the weights will be:

  • Class work: 10%
  • Homework: 5% times 5 homeworks (best of 6)
  • Paper project: 5%
  • Projects: 10% times 3 projects
  • Midterm: 15%
  • Final: 15%

Learning outcomes

At the end of the semester, you will be able to:

  • Analyze sources of error in numerical algorithms and reason about problem stability and how it influences the accuracy of numerical computations.
  • Choose appropriate numerical algorithms to solve linear algebra problems (linear systems, least squares problems, and eigenvalue problems), taking into account problem structure.
  • Formulate nonlinear equations and constrained and unconstrained optimization problem for solution on a computer.
  • Analyze the local convergence of nonlinear solver algorithms.
  • Reason about global convergence of nonlinear solver algorithms.
  • Use numerical methods to solve problems of practical interest from data science, machine learning, and engineering.

Course policies

Inclusivity and accommodation

We (Cornell as a whole, CS as a department, and I as the course instructor) are commited to full inclusion in education for everyone. Services and reasonable accommodations are available, whether you are facing permanent or temporary disabilities, immigration status issues, mental health or other personal challenges, or other types of learning challenges. If circumstances affect your ability to participate, let me know. Some resources that might be of use include:

Ingredients for success

To be successful in the course, I ask that you focus on

  1. Prepare. Course notes will be posted ahead of class, and we will provide pointers to supplementary reading. Come armed with questions!
  2. Engage. In class, we will ask you to ask questions and answer questions. This is part of your grade, and a chance for you to practice what you’ve learned (and help us understand when you’re confused).
  3. Start homework early. We want to help you figure out the homework, but to manage this we need time for you to get confused, ask us for help, and repeat a few times.

Late work policy

Unless otherwise stated, all work is due by 11:59 pm on the due date. All homework and projects should be submitted via the course management system (CMS); you are encouraged to submit early versions, since resubmissions up to the deadline are counted without penalty. For each assignment, up to three “slip days” are allowed. Over the semester, you may use a total of six slip days. You may not use slip days for the take-home midterm.

If you need additional accommodation, ask in writing in advance, with rationale and a plan for when you will be able to submit the work.

Collaboration

An assignment is an academic document, like a journal article. When you turn it in, you are claiming everything in it is your original work, unless you cite a source for it.

You are welcome to discuss homework and projects among yourselves in general terms. However, you should not look at code or writeups from other students, or allow other students to see your code or writeup, even if the general solution was worked out together. Unless we explicitly allow it on an assignment, we will not credit code or writeups that are shared between students (or teams, in the case of projects).

If you get an idea from a classmate, the TA, a book or other published source, or elsewhere, please provide an appropriate citation. This is not only critical to maintaining academic integrity, but it is also an important way for you to give credit to those who have helped you out. When in doubt, cite! Code or writeups with appropriate citations will never be considered a violation of academic integrity in this class (though you will not receive credit for code or writeups that were shared when you should have done them yourself).

Academic Integrity

We expect academic integrity from everyone. School is stressful, and you may feel pressure from your coursework or other factors, but that is no reason for dishonesty! If you feel you can’t complete the work on the own, come talk to the professor, the TA, or your advisor, and we can help you figure out what to do.

For more information, see Cornell’s Code of Academic Integrity.

Emergency procedures

In the event of a major campus emergency, course requirements, deadlines, and grading percentages are subject to changes that may be necessitated by a revised semester calendar or other circumstances. Any such announcements will be posted to the course home page.