SIAM Conference on Applied Mathematics Education, Invited Address
SIAM Annual Meeting, Portland, Oregon
Tuesday, July 10, 2018
11:45 AM - 12:30 PM
Since teaching is best discussed using examples (and I don't have any theories about teaching, except to be active and enthusiastic and conscious of your audience), I would like to report on my experience this year with a new second course on linear algebra. The course builds to an analysis of neural nets and deep learning. Any student is welcome to come. For this very active topic, undergraduates and graduates want to learn about it and many have seen examples. The major components are
We quickly had 100 students. It is an example of "computational thinking" and some of the homeworks are online labs --- not everyone would need to do this and it is certainly not my strong point. The course notes are growing into a book, because I believe math departments could and should offer such a course : Why should engineers have all the fun ? (They could respond that they created the subject -- but a lot depends on the SVD and vector and matrix norms and best approximation by rank k.)
I hope that this project will serve as an example for discussion of more active teaching.
Gilbert Strang
Massachusetts Institute of Technology, USA
Tuesday, July 10
This minisymposium combines the work of the Education Activity Group and the Machine Learning Activity Group. Deep learning refers to the revolution in artificial intelligence that has arrived with the construction of neural networks with many hidden layers. Deep teaching refers to the revolution that is needed — particularly in departments of mathematics and applied mathematics — to present this topic to students who want to learn about it. The background should include linear algebra and statistics and optimization. But not all students can wait for years and take these subjects first. When taught together they reinforce each other — especially if the course involves computing.
One goal is to enlist the audience in thinking about how courses in deep learning can be created — when the demand and the interest are already there for the students. It is a chance on a scale that doesn't come often! Our broader goal is to discuss new ideas and applications in deep learning and new ways to communicate to students.
Organizer: Gilbert Strang
Massachusetts Institute of Technology, USA
8:30-8:55 The Structure of a Deep Neural Net
Gilbert Strang, Massachusetts Institute of Technology, USA
9:00-9:25 Applications of Deep Learning in the Classroom
Loren Shure, MathWorks, USA
9:30-9:55 On the Role of Optimization Algorithms in Deep Learning
Srinadh Bhojanapalli, Toyota Technological Institute at Chicago, USA
10:00-10:25 An Informal Approach to Teaching Calculus
Pavel Grinfeld, Drexel University, USA