Making Deep Learning Human with Prof. Gilbert Strang (S1:E3)

Key Takeaways

Typically, the first few days of class, these guys ask, you know, "What's the class average going to be? How are we going to be graded?" I don't have any answers for that stuff. So, I say what is totally true, that I don't feel my main job is to grade them. My job is to teach them or learn with them. That's what I continue to do, and gradually they begin to believe. Today on the podcast, we're talking with teaching legend Professor Gil Strang. Maybe the key point is to make it human. You know, you're you're you're a person, like the student is a person. The book isn't quite a person, but it was written by a person. Welcome to Chalk Radio, a podcast about inspired teaching at MIT. I'm your host, Sarah Hansen, from MIT Open Courseware. One of OCW's most popular courses is Professor Strang's 18.06 Linear Algebra, a key foundation for his new course on machine learning, in which he's teaching students to teach computers. Professor Strang is known for inspiring students through his teaching. One YouTube commenter sums it up well. Quote, "This is not lecture. This is art." We wanted to talk with Professor Strang to see how he's been able to make complex math concepts engaging and accessible. We'll pick up our conversation with his explanation of what his new course, 18.065 Matrix Methods and Data Analysis, Signal Processing, and Machine Learning, is all about. So, this this is my adventure into the subject of deep learning. For example, recognizing an image, recognizing a zip code, a bunch of numbers, translating languages, or playing chess. So, that's what the course is about. How How does the machine learn? Essentially, the idea is say take an image then the the the deep learning system leads the machine to look at examples. We all learn from examples, the machine learns from examples. And by from from many examples or many chess games or many pages of of Chinese, you learn what's happening. And and what the the math part is is that the machine ultimately tries to assign a certain weight to a certain number big or small to each part of the image. So, perhaps if you're drawing a three, uh the image the computer recognizes a three, of course, by the curves and and gives less weight or zero weight sometimes to the empty space around the three, but picks out that three. So, that's the idea of deep learning. Traditionally, math courses have been defined by testing, which honestly makes sense. There's typically a right and a wrong answer in math. If you know the operation, if you do it right, you should get the answer. Tests can be a great vehicle for strengthening and measuring student skills, but Professor Strang's approach is different. So, I ask everybody to do a project. There is no final exam. Actually, there's no exam at all. I shouldn't like say this, but uh that's really what the subject is is having an idea of how Okay, I'll use deep learning for some thing. Like the recent proposed project was uh can you identify what makes a an image or or a picture attractive? Mhm. So, somebody has to say, you know, these pictures are attractive, these are not. We have to tell the computer something. What did that feel like to try something new pedagogically? Oh, it's fun. You know, I I like teaching and this is a subject where students just come from everywhere cuz they know it's stuff to learn and they've heard about it and they some of them know more than me but and then they those students write even better projects. Yeah, it's just so I do the lectures for the first three quarters of the course and then I try to get them to present. Which is a great experience for them so it takes a little urging to get them but yeah, yeah, it's really just wonderful. What insights have you gained about um having more of a student-led course and a project-based course? Any anything that other You realize slowly but finally that that's how people learn by by doing there that you couldn't give them a better way to learn than than create a project and usually it's on some topic they know about or they they're interested in like how do you find a criminal in a bunch of people? Yeah, it's just it's a very effective way to learn and something it's something that gets remembered where taking uh doing exam questions that I might make up sort of mathy questions. I don't know if that's remembered 10 years later but I think people's projects are. Along with this new approach comes a new paradigm for measuring student learning. Projects involve more than right and wrong answers. Projects are subjective and bringing the subjectivity into a math course comes with some initial skepticism especially from students who are so used to the typical learn the subject, perform on the test way of doing things. One of the things that makes Professor Strang and his courses so special is that he's not attached to these paradigms. In 18.065, in one of the videos, um you talk about grading students' work. And and you tell them that although this is important to grade their work, it's not your main concern. Your main concern is actually learning with them. Right. Um could you talk a little bit about that? So, typically the first few days of class, these guys ask, you know, "What's the class average going to be? How are we going to be graded?" I don't have any answers for that stuff. So, I say what is totally true, that I'm not I don't feel my main job is to grade them. My job is to teach them or learn with them. And uh that's what I continue to do, and gradually they begin to believe, you know, at at the beginning it's uh they still think, "Okay, but he's got to give me a B or a C or an A." Uh but uh really that's not what 18.065 is about, uh a grade. It's just not. Math is a subject you do, you don't just read, you you have to do it, you have to think about it. The way to learn math is to get into it and and uh Mhm. work on a thing which takes some thought. You can It's not You don't see it immediately, but you see it eventually. One of my favorite takeaways from Professor Strang's approach is that he centers his lessons around the humans in his class. For him, it's about engaging with the students in his course as people, and the learning is done by everyone. Well, first I'd like students and I want to help. And maybe the key point is to think with them, not to just say okay, here it is, listen. Listen up. I think through the the question all over again as they do. And and and you have to give time. You can't zip through a proof because class has to be sort of thinking with you and it happens that I lose the thread or I come up to a dead end where I don't know what I'm supposed to do next, but well, that's okay cuz students are going to hit dead ends, so it's seems to me it's okay for me to get stuck too and then give they see, oh, okay, maybe this this is the way to get out of that corner. I suppose I try to think it through once again and that then you sort of automatically see the word you you recognize what words you need to use and and what step what the steps are. Yeah, if you if you don't if you're not thinking it yourself, then you're probably going too fast and not connecting with the the thinking of the class. Of course, you don't know what everybody's thinking in that class. But uh overall, if if you stay sort of conscious of the class conscious of where they are, that's I think the thing for any speaker. Mhm. Is to be conscious of the audience and it's it's maybe the key point is to make it human. You know, you're you're you're a person like the student is a person. The book isn't quite a person, but it was written by a person. And to see that it's just like a natural thing to do. Mhm. Yeah. So essentially I think the thing is I like students. I like math and putting them together is just the best job in the world. Professor Strang shares additional thoughts on teaching linear algebra and matrix methods in data analysis, signal processing and machine learning in videos within the instructor insight sections of his OCW courses. You can find them at ocw.mit.edu. While you're there, download the teaching resources from his courses and watch his lecture videos. Discover the magic of his teaching for yourself. We're so happy to bring you conversations with MIT faculty who are passionate about impacting the world in positive ways. Write to us to share your story of how you're using OCW materials to shape your world or those of others. Until next time, I'm Sarah Hanson from MIT OpenCourseWare.