A Practical, Hands-On Approach to Machine Learning Education
If you enroll in a university-level data science program, you’ll likely spend months learning machine learning theory and how to code them. But you won’t necessarily see how they relate to real-world problems. That’s what spurred the folks at Dato to work with Coursera and the University of Washington to create a class that advocates a more practical approach to machine learning.
As the CEO and co-founder of Dato (formerly GraphLab) and the Amazon Professor of Machine Learning at the University of Washington, you could say that Carlos Guestrin knows a thing or two about machine learning. But one thing that’s bugged him is how machine learning gets taught in schools.
“In data science classes, you’re always focused on the theory or the algorithms [but] you’re not thinking about the practical aspects of how machine learning gets done, such as how it gets evaluated, how you interact with your data, and how you understand if it’s working or not working,” Guestrin tells Datanami. “Those questions only get addressed in practice as you’re doing a job for a particular project.”
Guestrin hopes to remedy that shortcoming with a new online class offered through Coursera in conjunction with the University of Washington. The six-course specialization, which starts Monday November 9 and costs $474, is designed to provide a more real-world approach to teaching machine learning concepts and techniques.
By starting at the end, so to speak, Guestrin hopes to set the stage for what’s possible with machine learning, and hook the students’ imaginations. “Unlike typical machine learning courses, where we talk about what’s the probability and proofs of thermos from the algorithms, we start with high-level use case,” he says. “What can you use it for? Then by going through the use cases, you understand the mental process and methods you need to build applications that use machine learning.”
Because the class is offered online and is self-paced, it should attract a broader audience, says Emily Fox, who is an Amazon Professor of Machine Learning and Assistant Professor in the Statistics Department at University of Washington, and who will be teaching the class with Guestrin.
“Our goal is to avoid the standard prerequisite-heavy approach used in other ML courses,” she writes on a Dato blog. “Instead, we motivate concepts through intuition and real-world applications, and solidify concepts with a very hands-on approach. The result is a self-paced, online program targeted at a broad audience and offered through Coursera with the first course available today.”
Only after learning about the use cases do the professors get into some of the nitty gritty details of machine learning. Over the course of several months, students will take classes on regression algorithms, classification algorithms, clustering and retrieval algorithms, and building recommender systems and dimensionality reduction. The final class “Machine Learning Capstone: An Intelligent Application with Deep Learning,” starts in April.
Students who take the class will become familiar with GraphLab Create, the commercial machine learning product from Dato that’s used by companies like Zillow, Pandora, and StumbleUpon. The company recently added several machine learning toolkits to the product that provide pre-built functionality in the areas of recommendation engines, image search, churn prediction, and sentiment analysis.
As a professor, Guestrin sees the lack of trained data scientists as a problem for organizations hoping to take advantage of the power of machine learning to improve their offerings and processes. But as a CEO, he sees that as an opportunity.
“The bottleneck today is the lack of qualified talent on the one hand and the lack of tools to accelerate the creation by requiring less talent,” he says. “Today you have to do most things from scratch. There’s lots of open source tools, but that doesn’t give you a production service. You have to really understand how things work, understand the data and the math behind it, and understand the largescale distributed systems that you have to interact with to take that math into reality.
“That part is extremely slow and painful and only a few people have the expertise to do it,” he continues. “The challenge and opportunity is to figure out how to make these things easy and robust so the wide variety of people who are really excited about building those intelligent applications can go ahead and do it.”
There’s still time to enroll in the six-course Machine Learning Specialization at Courseara. For more information, see https://www.coursera.org/specializations/machine-learning
(feature image by Sarah Holmlund/Shutterstock.com)