Why use the LMS for linear systems?

It’s been a bit of a whirlwind since the last post but I made my course website and “published” it. Rutgers has basically forced all courses into their preferred “Learning Management System” (LMS) Canvas. Even the term LMS has some weird connotations: is it a management system for learning or a system for managing learning? A system for students to (barely) manage to learn? Canvas in particular seems terrible for things math-related (one semester the entire LaTeX rendering engine crashed with no notice) or engineering-related, and in general the whole question management system is garbage.

So if Canvas is so awful, shouldn’t I use something else? Maybe. It helps to imagine (with some dramatic liberties) the evolution of the course website:

  • Everything’s on paper. There’s a book or lecture notes/a reader you buy from the bookstore or copy shop, assignments are physical handouts only (photocopies or dittos or something). Every class works this way. Scores for assignments have to be manually associated with students.
  • Same book or lecture notes/reader but homework files are on the web (or ftp or gopher maybe) in .ps (or later .pdf). A course website is some hand-coded HTML (like my current homepage!) Students can maybe pick up a printout or print it themselves in a computer lab or at home. We can call this the “bag of PDFs” model. Scores for assignments have to be manually associated with students.
  • The website design is somehow centrally controlled either via a template or something and now the book/notes are in .pdf but many students still print things out because lugging a laptop around feels annoying. Maybe a pretty bag of PDFs.
  • The dawn of the LMS: student rosters can get associated into a system where you can deposit your bag of PDFs and then organize them in some pre-specified way. Grades for assignments can be manually associated with students in the system and then you can submit them automagically!
  • The creep of the LMS: you can make quizzes/assignments (simple ones) that are auto graded, make your site look pretty, maybe embed some videos and provide other content, and generally automate some aspects of your class. To take advantage of “features” you have to change your class to meet the tools. The latter is appealing because some features help students learn better (at least according to some research). It’s an opportunity to try things new.
  • Late-stage capitalism LMS: Universities “mandate” that faculty use a particular LMS. Many faculty comply. Every year new edtech companies show up trying to get you to use their software (and sometimes steal student data). Some might be grifts, others are not. They heavily marketed and different ones are pushed by different teaching and learning centers. Many of them require you to change your teaching to fit the tool because they are one-size-fits-all: claims to works for all fields, all types of classes!

So why am I still using Canvas? The main reason is that it benefits the students, not because it is good, but because they have been using it for 2 years and they are used to it. They can see all their due dates on a single dashboard. If you are reading this and scoffing, think about how terrible and non-interoperable almost every calendar system is. If you say “you’re just bowing to peer pressure” you’re more or less right. If you say “but it builds character for students to have to manage their due dates” then I’d ask if the goal of your class is to teach the material or to teach time management. If you say “both” then does your class explicitly teach students time management skills? I’m guessing not.

In this new class format, there are 28 class sessions. In 26 of them there is a conceptual quiz students have to take before class as well as an in-class assignment for which they have to upload solutions. Then there are homeworks, quizzes, and projects/labs. Compared to the old 8 problem sets, two midterms, and a final, that’s more than a 6x increase in things to keep to track of for > 200 students. It behooves me, as someone who cares whether students learn the material, to try and make keeping track easier.

Ultimately a university “adopting” any LMS is coercive because if students use it for all their introductory classes then using something else is almost deliberately making their lives harder with no real benefit. I don’t think I’m going to William F. Buckley it up and stand athwart with my bag of PDFs (even if it is on github). Ultimately I think using the LMS the right thing to do by the students. I’m going to be super salty about it though.


An experiment in teaching Linear Systems and Signals

This fall I am teaching for the n-th time our introductory signals and systems course (ECE 345). This time I’m teaching all of the students (at the time of writing, 206 of them): prior offerings split the class into two sections taught by different faculty and in the last two years of COVID-induced remote instruction I co-taught a combined class with my colleague Salim El Rouayheb. I had thought about some plans to change a few things about the class based on my last in-person offering in 2019 and drew up a few ideas, planning to get things organized in the month before the semester.

At the beginning of August I went to campus to look at the classroom, which is Lucy Stone Hall on the Livingston campus at Rutgers, which can seat 400 students. For context, to get there from my office/where most STEM classes are (on Busch campus) one has to drive, take a bus, or bike. I previously taught in a classroom that can seat 147 which had whiteboards on rollers that went up and down (they are behind the projection screen), which gives 4 boards worth of space visible at a time. The new classroom supposedly has “5 chalkboards” but those are on wooden panels that are partially obscured by the podium. The chalkboard on casters shown in the picture is not particularly visible from the back of the class. So… no real board space. I can use the projection screen with a tablet of document camera (or maybe transparencies to be really old school) but with a single projection screen.

So I’ve embarked on a far too ambitious plan to partially “flip” the class: students will watch video lectures (already recorded during COVID times) and then come to class to do more active learning/problem solving activities. Since this blog has been moribund for the last few years, I will try to write about this process as it goes to help process/document what I’m doing and how well its working (or not).

I’ve been doing a bit of reading on prior approaches, including:

I’ve also gotten a lot of help from various friends and other educators about their own experiences and ideas of what has worked and what hasn’t.

I rapidly realized that I could not implement all of these ideas in a month before the semester so I am trying to pick and choose my battles. Usually, to successfully flip a class requires a high instructor (TAs, learning assistants (LAs), etc.) to student ratio. I don’t even know how many TAs I’ll have this semester yet, so that’s going to be a challenge. It’s waaaaay too late to ask for LAs. Hence I’m calling this a “partial” flip.

I’m not sure I’ll be able to pull it off, but here’s hoping!

Teaching students to stay away from Physiognomic AI

I read Luke Stark and Jevan Hutson‘s Physiognomic AI paper last night and it’s sparked some thinking about additional reading I could add to my graduate course on statistical theory for engineering next semester (Detection and Estimation Theory).

“The inferential statistical methods on which machine learning is based, while useful in many contexts, fail when applied to extrapolating subjective human characteristics from physical features and even patterns of behavior, just as phrenology and physiognomy did.”

From the (mathematical) communication theory context in which I teach these methods, they are indeed useful. But I should probably teach more about the (non-mathematical) limitations of those methods. Ultimately, even if I tell myself that I am teaching theory, that theory has a domain of application which is both mathematically and normatively constrained. We get trained in the former but not in the latter. Teaching a methodology without a discussion of its limitations is a bit like teaching someone how to shoot a gun without any discussion of safety [*]

The paper describes the parallels between the development of physiognomy and some AI-based computer vision applications to illustrate how claims about the utility or social good arguments made now are nearly identical. They quote Lorenzo Niles Fowler, a phrenologist: “All teachers would be more successful if, by the aid of Phrenology, they trained their pupils with reference to their mental capacities.” Compare this to the push for using ML to generate individual learning plans.

The problem is not (necessarily) that giving students individualized instruction is bad, but that ML’s “internally consistent, but largely self-referential epistemological framework” cherry picks what it wants from the application domain to find a nail for the ML hammer. As they write: “[s]uch justifications also often point to extant scientific literature from other fields, often without delving its details and effacing controversies and disagreements within the original discipline.”

Getting back to pedagogy, I think it’s important to address this “everything looks like a nail” phenomenon. One start is to think carefully even about the cartoon examples we use as examples. But perhaps I should add a supplemental reading list to go along with each topic. We fancy ourselves as theorists, but I think it’s a dodge. Students are taking the class because they want to learn ML because they are excited about doing machine learning. When they go off into industry, they should be able to think critically about whether the tool is right for the job: not just “is logistic loss the right loss function” but “is this even the right question to be asking or trying to answer?”

[*] That is, very American?