CISC371, Nonlinear Optimization: Fall 2021

Calendar Description:

Methods for nonlinear data analysis, particularly using numerical optimization. Applications may include: unconstrained data optimization; linear equality constraints; linear inequality constraints; constrained data regression; constrained data classification; evaluating the effectiveness of analysis methods

Lecture and Report Materials:

The navigation panel, on the left, leads to a list of the lectures. These are video recordings and notes for the course. Guides to writing reports, for students and for graders, are also on the left.

Lectures:

In these unsettled times, the instructor realizes that a student may have mixed feelings about congregating densely in classrooms. These feelings might change through the semester and there is always a possibility that classes need to switch to remote delivery. Recent experience has shown that some material is, actually, better delivered remotely; an example is carefully prepared lectures on complicated topics. At Queen's, we recognize that the communal experience of a classroom is irreplaceable, as are the human connections we develop outside of the classroom. There may be a solution to the conundrum of choosing between purely in-person classes and purely remote classes.

This course is given as balanced classes. The lecture material for each week will be given before the week begins, in the form of video lectures and course notes. Each student is expected to view the videos and read the notes before attending the scheduled class times.

Classes are, effectively, "super-tutorials". The teachers are the instructor and the Head Teaching Assistant (Head TA). One teacher will appear in person during classes that are not due dates; on these dates, the other teacher will appear remotely in a session that will be recorded for later review by students. No classes will be held on days when assignments are due or when tests are held.

The effect of a balanced classroom is that, each day, a student can choose to attend in person or remotely. The same syllabus and teaching examples will be used; the only difference will be the questions and responses that spontaneously arise during the class. This way, both in-person and remote students will have an optimized experience for learning the material. Students will be neither penalized or advantaged by their choice of how to access each class.

Brief Description:

When we use machine learning or data analytics methods on structured numerical data, we often want to find an optimal answer. For example, if we want to model the data as being linearly related, we might want to minimize the error between the model and the data.

A guiding principle for this course will be that linear algebra is our main way of describing and implementing nonlinear analysis of structured data. Most other courses use a "calculus-first" principle; instead, we will use a limited amount of calculus to derive the linear algebra that we need for writing code. This "algebra-first" principle will allow us to explore complicated ideas in a concise and scalable manner by using linear algebra as the language of structured data.

Further details can be found from the "Description" link, to the left.

The course contents, including the notes and supplementary material, are provided in the onQ learning management system.

Queen's University is situated on traditional Anishinaabe and Haudenosaunee Territory.


Last updated