Classroom Q&A: Dr. Craig Martell and Machine Learning

Machine learning enthusiast and expert Dr. Craig Martell has been a Khoury College lecturer since 2018, teaching courses in algorithms and machine learning. His fascination with the mathematics behind computing led him to co-author Great Principles of Computing. As head of machine learning for Lyft, Dr. Martell breaks down machine learning in this interview with Khoury News, where he discusses his course, his industry experience, and his love for Seattle’s landscape.

Dr. Craig Martell

Tell us about your background, academic and professional.

I like to say I had a misspent youth. I knew I wanted to get my PhD, but before figuring out I wanted to go into computer science I enrolled in several master’s programs. I got master’s degrees in philosophy and political science. I then made a switch to logic, but I didn’t get another master’s because I already had too many. I finally finished with a Ph.D. in Computer Science at the University of Pennsylvania. My goal in moving into computer science–AI in particular–was my fascination with humans (which is the opposite of why most people go into this field). In computer science I was interested in what I call “testable philosophy”: specifically, can we design systems that would allow us to understand how humans behave?

The bombings on 9/11 happened while I was in graduate school. My plan was always to be an academic, and that didn’t change. I just focused my academic goals towards things I thought could help with the world after 9/11. So, I accepted a job as a professor at the Naval Postgraduate School (NPS). All of the students were naval officers or military officers, and the research directly helped the Navy. So, I was a professor for about twelve years. Then in 2013, I decided to spend some time in industry. Machine learning was starting to take off and my skills were in demand.

How did you get to Khoury College?

I started teaching again, at Northeastern, because I missed being in the classroom. As soon as my family and I moved to Seattle, I saw the Align program, was really motivated by it, and contacted the director about teaching.

Tell us about your course, Machine Learning, what do you cover?

Machine learning is the AI that works. AI had a lot of promise for a lot of years and never really got very far. Machine learning is the stuff that’s actually been really successful. Machine learning is basically statistics at scale. We count all kinds of features under different contexts from the past. And we use those features in similar contexts to predict the future. For example, when you type a search query, the system will return documents that are like the documents that people who have typed similar queries have clicked on in the past. We use all of that past information to make a guess, to make predictions.

In the course, I cover the basics: How do you gather the data? How do you evaluate it? How do you know the statistical model is good? I would say ten years ago, we would have covered ten or fifteen different techniques, but almost all of those techniques have been superseded by deep learning. The first month of the class is about the foundations that you need to be able to understand the second half of the class, which is all about deep learning, and how deep learning applies to language, vision, speech, etc.

Why is it important for students to learn about Machine Learning?

Machine learning systems already have a very large impact in your life. And, if we are not careful, that impact could be scary, biased, and/or dangerous. As a software engineer, you will need to understand how machine learning works – both because more and more general software engineers will be building machine-learning systems, and so you can ask the right ethical questions about the systems you are being asked to build.

Secondly, I think all citizens should learn some aspects of machine learning; because it already impacts everything from the jobs that you’re being offered to how the police interpret you through facial recognition. I was on a panel this year about bias and machine learning for this great documentary called Coded Bias. It’s essentially about how facial recognition hasn’t worked if you’re female or if your skin is darker. It traditionally has worked really well for white men. This bias is because the data used to train the system was gathered using the researchers at the time – who were mostly white men. These algorithms are already running in the world. And if we’re not prepared as citizens to understand how they work, we can’t argue against them when they get it wrong.

What else do you teach for Khoury College and at what level?

Last summer, I taught the algorithms class. I’ll probably teach that again. I really like algorithms; it is essential to computer science. I also teach CS 6140: Machine Learning, CS 8674: Masters Project, and CS 7140: Advanced Machine Learning

Besides your work at Khoury, what else do you do professionally?

I run machine learning for Lyft, and that’s a pretty cool role. It’s not just about creating new algorithms, although that’s a big part of it. It’s also about creating the infrastructure that allows those algorithms to run, allows those algorithms to be evaluated, and allows us to ask questions like, “Is this algorithm delivering against the business goal it was built for?” I think there’s a point of inflection in industry now—a realization that the model by itself, which might solve some really cool, sexy problem, is now slightly less interesting. What’s more interesting is answering questions like: “Is it actually achieving the business goal?” Or “Is it running efficiently?”

This last question is becoming more and more important. Large AI models can produce so much carbon that they’re actually bad for the environment. A lot of what my team at Lyft has shifted to isn’t just about creating the sexy machine-learning model, but it’s about operational excellence surrounding the model.

What topics in computer science interest you? Do you have any specific problems or applications that you’re particularly passionate about?

I’m kind of in love with two sides of computer science. I really like machine learning. On the other hand, I also like operating systems, the aspects of CS close to the “metal” itself. In one sense, these are very far apart. Machine learning is abstracted from the underlying computational properties. It’s mostly about how humans interact, or about how we can predict human behavior. On the other hand, I like the stuff at the bottom, like algorithms, which is core to computer science. I’m also fascinated by complexity theory, which is the math that’s core to computer science. In the book I co-authored for MIT; we cover a lot of those kinds of concepts.

What’s it like teaching at the Seattle campus? What do you like most about Seattle?

Great skiing is an hour and a half away. I grew up in Vermont, and I love to ski. I’m teaching my four-year-old to ski this year, and my nine-year-old is already very good. In Seattle, good skiing is 40 minutes away, and excellent skiing is only an hour and a half away. The natural beauty here is just unbelievable.

What’s a fun fact about you, that your students might not know?

I have a black belt in Taekwondo, and I have an Erdős number of three.

Subscribe to the Khoury College newsletter