Answers From The Instructor: Mathematics for Machine Learning
When building a career in AI, it makes sense to learn from the best. That’s why we asked Luis Serrano to teach our latest specialization, Mathematics for Machine Learning and Data Science.
Luis is a lifelong mathematician and a machine learning scientist by trade. He has taught courses on probability at Université du Québec à Montréal and the University of Michigan, built algorithms for Google, taught AI at Apple, and currently works on language models for Cohere. His YouTube channel, Serrano Academy, is known for its approachable, bite-sized explanations of a variety of math and machine learning topics. And somewhere amid all of that, he found time to write a book: Grokking Machine Learning.
Luis recently spoke to us about his own learning journey, what people can expect from the Mathematics for Machine Learning and Data Science Specialization, and his secret to making mathematics approachable to anyone.
Have you always been good at math?
I was terrible at math until Grade 8. I failed at it because I couldn’t relate to the lessons, which were full of formulas and high-level concepts. One day, however, I took a national math exam. The test didn’t have many formulas. Instead, it asked you to solve puzzles and play math games. I remember thinking it was fun. When the results came out, it turned out I had ranked first in the country where I grew up, Colombia. I realized that, not only did I like math, I was actually good at it.
How did you first get interested in AI?
About ten years ago, I was a math PhD who was teaching and doing research at the Université du Québec à Montréal. I enjoyed the teaching and I found the research interesting. However, I was curious about other topics and I wanted to do more applied work, so I started programming. This led me to machine learning. I took Andrew Ng’s course and realized right away that it had all the math I loved, but applied from a slightly different angle. I was hooked.
What is the story behind your YouTube channel, Serrano Academy?
After my fellowship at the University ran out, I decided to switch careers. I got a job at Google, where I built machine learning recommendation models for YouTube. It was a lot of fun, and that hands-on research really taught me how to apply my math skills to AI. During the weekends, I was doing workshops and lecturing. I came to realize that I loved my side hustle as a teacher so much that I wanted to make it my full-time job.
I left Google and got a job at Udacity, where I was in charge of teaching the machine learning course. When I started teaching courses online, I had to share a lot of material with my peers. This was done in order to get feedback, and also to show my team the directions I was planning for the course. As a way to speed up this process, I started making my own videos presenting various course topics and sharing them around the company. For instance, if I wanted to explain linear regression, I would make a five minute video and send it over email. Then I would get my colleague’s reactions, as a sort of alpha test, and use that to fine-tune my delivery.
One day, I was teaching several topics and made a 30 minute-long video. It wouldn’t send over email, so I posted it to my personal YouTube channel. Remember, I had worked at YouTube, so I was aware of how difficult it is for random videos to gain traction. I assumed that only my coworkers would watch it. However, it wound up getting widely shared and things took off from there. That original video now has almost a million views.
How do you relate to beginners who take the Mathematics for Machine Learning and Data Science Specialization? Especially those who are being exposed to a math or machine learning concept for the first time?
I’ve always been a very slow learner. I struggle when thinking about concepts at a high level and have to get to know them through real-world examples in order to understand. Even today, I have trouble keeping up in conversations with other mathematicians and scientists, because many of them tend to speak in more abstract terms. So, not only do I know what it is like to struggle with learning difficult topics, I also think in terms of real-world examples and analogies. That comes through in the way I teach, and I believe it resonates with a lot of learners.
Who is your target audience when you are teaching a topic?
I have a one-size-fits-all approach to making videos. I want to engage beginners by exposing them to new ideas in ways they can grasp. I also want to hold an expert’s attention and have them think, “Oh, I didn’t know that,” or, “I’ve never thought of it that way.” In short, I try not to bore the experts and I also try not to lose the beginners.
How would you describe your teaching style?
My lessons are hands-on and contain lots of graphics and examples. I also try to make them as much fun as possible. When I’m crafting my examples, I try to piggyback on things that people already know. Surviving in the real world requires a lot of logic. I find that most people are very good at math, but they don’t know it, because they’ve only ever been taught using formulas. By basing my lessons on their real world experiences, I am able to gradually introduce them to new topics without any confusion.
How does the Math for Machine Learning and Data Scientists Specialization differ from the lessons you offer on your YouTube channel, the Serrano Academy?
If you were to watch 30 seconds from my channel and 30 seconds from the specialization, you would see the teaching style is very similar. The difference is in the larger structure. At the Serrano Academy, I have a large collection of videos, and each is about its own small topic. In the specialization, I am telling a complete, interconnected narrative about mathematics. My YouTube videos are like short stories, and the specialization is like a novel.
The other component that makes the specialization different is all the stuff the development team did to make learning more interactive. You will have access to tools that allow you to visualize concepts as you learn them and labs to help you practice what you have just learned.
Andrew Ng is famous for saying “Don’t worry about it,” when it comes to math. Why do people in machine learning and data science need to know math?
I agree with Andrew. You don’t have to know math in order to break into machine learning or data science. It’s like how you don’t need to know how a combustion engine works in order to drive a car. However, if you want to drive a Formula One car, you need to know some things about the vehicle’s mechanics because that will help you optimize its performance.
The same is true with machine learning. You don’t need to understand math in order to build and deploy models or interpret their results. However, an understanding of mathematics will enable you to tweak your models and optimize their performance. Furthermore, I don’t believe that students have to go learn math and then come back to learn machine learning. They can use what they know to start building models and analyzing datasets, and learn the math as they do machine learning. That way it is more fun and hands-on. They will also remember it better since the lesson is attached to the experience of solving a real-world problem.
What makes the Mathematics for Machine Learning and Data Science Specialization different from a typical math course?
First, our team has worked really hard to make this course fun, with graphics, examples, and interactive animations. Second, we tie everything we teach in the course back to its practical application in machine learning with real-world contexts and applications.
Other math courses tend to drop learners straight into the deep end of math. They want you to memorize formulas then apply them blindly. Our approach provides learners with a real-world example for every concept so they can clearly see how it should be applied. And we give them interactive exercises and tools so they practice what they learn in real time.
What skills or prerequisites should people have before taking the specialization?
All you need to know is basic high-school algebra. If I say x+1=7, you should be able to tell me that x equals 6. You should also be familiar with equations such as the quadratic equation.
You should know how to plot a function on a graph. For instance, what line corresponds to the equation y=x?
You should understand the basic concepts of probability. For example, if I tell you that I flipped a coin and it landed on heads seven times out of 10, you should be able to infer that the probability that the coin landed on its head is 0.7, or 70%.
Do you need a background in machine learning or data science to take the specialization?
No. This can be taken as a standalone math course for anybody, in any career. We just happened to pick a bunch of math topics that are useful for machine learning. We also use machine learning to color our examples and provide real-world context, but these examples will be clear to anybody who has ever used, for example, an Excel spreadsheet.
How do you keep learning?
I live my life by one mantra: Learn every day. I have based every full-time job I’ve ever taken on that. For example, I jumped into quantum-computing research because I was curious. It was difficult, and I was lost much of the time in the beginning, but I enjoyed it because my assignment was to learn. I also learn in the process of teaching. For my YouTube channel, I often seek out a complicated topic that I do not understand. Then I challenge myself to find the simplest example that will explain how it works. Then I get to share that newfound knowledge with the world.