“Even though I really love music…after a while, I noticed there was a math- and science-shaped hole in my life,” Lipton says in describing his transition from jazz musician (you can find some of his music on YouTube) to machine learning and analytics academic.
You may not think to look for predictive analytics algorithms and lines of code among the flickering candlelight of a jazz bar. But there isn’t as much distance between data science and jazz music as you might expect. In fact, CMU Tepper School of Business professor Zachary Lipton holds passions for both, at least when he’s not founding new start-ups or teaching machine learning courses in our online business analytics master’s degree program.
Getting Jazzed About Machine Learning and Predictive Analytics
Jazz pianist Dave Brubeck once famously said, “Jazz stands for freedom. Get out there and improvise, and take chances, and don’t be a perfectionist – leave that to the classical musicians.” For Lipton, freedom came from the ability to pursue his intellectual curiosity and scholarship on a full-time basis.
“One of the great luxuries of being an academic is that you have a relatively broad license to research and explore a wide range of disciplines if you choose,” Lipton continues. “Over the course of my academic career, I’ve been able to research topics including machine learning methodologies, applied machine learning with healthcare applications, ethical concerns with automated systems and much more.”
One of the qualities that Lipton has kept from his days in jazz clubs is a tendency to challenge convention. Where jazz surprises and delights with musical improvisation, Lipton does the same for data science by creatively applying machine learning techniques to solve industry-wide problems.
Finding the Signal Through Noisy Data and Machine Learning Hype
When the machine learning community and many in tech media got a little overly enthusiastic in response to an advancement in natural language processing (NLP) from OpenAI, Lipton’s insight provided a measured perspective: Yes, it was a step forward, but it was a step on a path the rest of the artificial intelligence community was on already.
Skepticism is a healthy quality for analytics professionals. Machine learning algorithms and predictive analytics offer untapped opportunity, but as research firm Gartner points out, the hype around it can sometimes make it difficult to find the value for IT and business leaders. Creativity and innovation help push machine learning further, but only when guided by a purpose and a clear understanding of the goals driving the application of the technology.
This is one of the areas where professor Lipton’s perspective has proven invaluable. His background as an economist has helped him to hone-in on the practical side of predictive and machine learning technology.
Using Machine Learning to Find Answers in Medical Data
As natural language processing and machine learning were rapidly gaining traction, Lipton saw potential for the technology that few, if any, others had noticed. His research, published in 2015, became the first to use a specific technique for leveraging a machine learning algorithm in historical health information.
“When I started working in Deep Learning in 2014, a lot of people were getting excited about using sequential models for mining patterns in text data—people started using recurrent neural networks to develop automatic translation and speech recognition systems,” Lipton says. “When I was working in this area, I was more interested in applying a similar approach to finding patterns in medical data.”
The model Lipton and other researchers developed was able to classify 128 different diagnoses when given 13 clinical measurements. His work has since inspired more than 100 other publications and continues to serve as a foundation for exploring other machine learning applications.
“I forged a collaboration with David Kale, a Ph.D. student at USC, and data scientist affiliated with Children’s Hospital Los Angeles. Together, we got access to multivariate clinical time series data,” Lipton explains. “Basically, this type of data can be messy because it’s often recorded in different formats, on different time scales and with varying levels of completeness. So, traditional analytics methods at the time struggled to gain insights from it. We were the first to apply modern recurrent neural networks to this type of data for the purpose of recognizing diagnoses associated with specific patients. Our technique significantly outperformed other methods, and our work opened a promising new line of research.”
Where a lot of jazz music is about finding the right fusion of different styles, analytics is often about finding the right fusion between technology tools, business goals and data. Using the wrong method for the job results in conclusions that fall flat (or worse, conclusions that aren’t accurate at all). Finding the right approach, however, can incite applause in the conference room.
The Transition from Music to Machine Learning
In the first year of earning his Ph.D., Lipton balanced playing saxophone four nights a week while taking classes and studying. It didn’t take long, though, before his focus shifted and he devoted more of his attention to his passion for finding answers in data.
“There was this kind of huge explosion in interest for machine learning right around the same time as I was beginning my Ph.D.,” Lipton says. “As the field took off and I was finding success in it, I started spending more time focusing on the technology side of my career. I spent the summer of 2014 living in India. Then the following summer, I was living in Seattle and worked as part of a core machine learning team, helping to build the technology that went into Amazon’s recommendation engine for products like Prime Video. The year after that I worked in Microsoft Research. In my last year, I jumped into simultaneously finishing my Ph.D. as a full-time data scientist and working on a really talented team with Amazon AI.”
The explosive rise of industry-wide interest for machine learning, neural networks and artificial intelligence helped to propel Professor Lipton’s passion as well as his career. He continues to make contributions in machine learning, from work in named entity recognition to critically analyzing trends in the academic machine learning community.
Despite his busy academic life and career as a technology entrepreneur, Lipton does still find time to play music as a hobby.
“I still practice a bit, and recently, we got my old proper keyboard in our apartment,” Lipton says. “My girlfriend, who’s a much better musician than I am and a composer, gives me some pointed looks when I miss a key signature. Music has filled an important part of my life, but I have to admit at this stage that I’m in all-in on machine learning.”
Bringing Machine Learning Together with Predictive Analytics
Similar to the incredible growth in machine learning and analytics as a whole, investment in predictive analytics tools is set to increase by leaps and bounds.
In the same way that data science is a broad field that often uses machine learning, predictive analytics can encompass a vast array of technologies, tools and methods. When applied correctly, though, machine learning can help to create or tailor predictive models more quickly and solve problems that would otherwise be impossible.