John Hallman: machine learning and control theory

Wednesday, Jul 29, 2020

As the name suggests, automatic control concerns the control of dynamic continuously operating systems such as the cruise control on a vehicle, auto pilots on aircraft, industrial process control (paper, steel, chemicals), and building heating. John Hallman ’20 was intrigued by the concept of ``learning’’ how to control such processes and made it a focus of his senior year research.

In his independent project for the Center for Statistics and Machine Learning’s (CSML) certificate program, Hallman, a math major, noticed that many studies in the theory of control dealt with the setting full-information when various parameters and variables of the controlled system are assumed to be known. He decided to look at control in a different way by studying it in the context of bandit feedback, a situtation when many of the variables that drive a system are unknown.

Hallman developed a machine learning algorithm that deals with control in bandit feedback settings. The algorithm figures out the unknown variables in these systems. His project, Non-Stochastic Control with Bandit Feedback, shared the best poster award at this year’s CSML poster session along with the project of Florence Wang ’21.

“If you don’t know what the system is really doing, that’s the setting where my control algorithm works best,” said Hallman.

Hallman thinks that control in bandit feedback is an interesting issue to explore because it deals with the theoretical applications of algorithms – something he loves to study. The study of bandit theory is well established in the optimization field, but it has not been been a central focus of control theory. Hallman added that in practice you can put sensors in many places in order to eliminate unknown factors. Hence his algorithm is probably not of immediate interest to practitioners.

In addition to his CSML certificate, he also completed the certificate for the Program in Applications of Computing.

After graduation, Hallman joined up with a San Francisco start-up called Sisu, a data science company that serves businesses and industry, to take up a machine learning engineering position. What’s intrigued Hallman about the start-up is that the company strives to automate data analysis. Simply put, you connect your data base to the company and it does the analysis automatically and delivers findings back to you. In this way, data analysis becomes available to more organizations and more people who may not have a doctoral degree in data science, he said. In the future, the kind of systems the company produces will be more robust, powerful and simpler to use, he said.

“I’m really excited about this opportunity because I’ve been thinking about data analysis automatization for a long time,” he said. “I want to find ways to write good algorithms and to make data useful for many people in the real world.”

Hallman, who’s from Sweden, has been interested in math, computer science and machine learning for a long time. Hallman’s love for these subjects started in high school when he took classes with a teacher, Andreas Bergholtz, who pushed his students and didn’t allow them to be complacent.

“Before then, I really didn’t think of math as a field of study or a career,” he said. “But our math teacher was outstanding. He changed my life. Without him, I wouldn’t have found my passion. I became interested in problem solving.”

Hallman went onto compete in math competitions and won awards such as the silver medal in the International Mathematical Olympiad.

Hallman is also is a skilled figure skater and competed seriously in the sport while growing up. He won second place in the Nordic Junior Figure Skating Championships and in the Swedish Junior Figure Skating Championships.

After hanging up his skates and moving to the United States for his studies, Hallman focused his energies on math and machine learning. During his time at Princeton, Hallman was a summer research intern for Professor Elad Hazan at the computer science department. As an intern, he performed research on optimization and research learning. He was also a student research at the Google AI Lab at Princeton, where he delved into machine learning research and helped build out the machine learning infrastructure. He also won a student computer science teaching award.

As for future plans, he may get his doctoral degree but he wants to get some industrial experience first, Hallman said.

“I’ve been interested in the theoretical development of algorithms and their practical applications,” he said. “There is an enormous step between machine learning research and making these tools suffciently robust and simple that people can pick up the tools and use them.”