Puthawala: Deep learning expert deepens relationship with Japanese colleague

Mathematicians like Michael Puthawala are few and far between. Thanks to a National Science Foundation program, the SDSU assistant professor was able to get better acquainted with a similar mathematician 6,000 miles away.

Through a joint program between the National Science Foundation, the primary U.S. agency for funding fundamental science research, and the Japanese Society for the Promotion of Science, the counterpart agency in Japan, Puthawala was able to spend the first six weeks of this year in Japan. 

He spent extensive time with colleague Takashi Furuya of Doshisha University and was able to meet future potential collaborators as well as open doors for potential regular visits.

Puthawala joined SDSU four years ago after completing a three-year postdoctoral fellowship at Rice University in Houston. He is the CAPITAL Services Scholar in Artificial Intelligence and Machine Learning.

Originally from Binghamton, New York, Puthawala has a bachelor’s degree in mathematics from Rensselaer Polytechnic Institute (2014) and master’s and doctoral degrees in applied mathematics from the University of California, Los Angeles (2017 and 2019, respectively).

He has supplemented that with summer internships at Massachusetts Institute of Technology Lincoln Lab in Lexington, Massachusetts, in 2013-14; Oak Ridge National Laboratory, Oak Ridge, Tennessee, in 2016-17; and at Google in Los Angles in 2018.

NSF proposal targets Japan

The Jan. 4 to Feb. 14 trip came about through a request for proposals from the NSF’s Office of International Science and Engineering through its Division of Mathematical Sciences. The request specifically targeted Japan. The NSF doesn’t have a formal agreement with Japan, and Puthawala sees his trip as an early step in that direction.

The fact that he was based at Doshisha University is also perhaps a harbinger. Doshisha means a community created by those who share the same aspiration. 

The aspiration that Puthawala and Furuya share is research into the fundamentals of deep learning, which is the mathematics that powers all machine learning. “Chat GPT and automatic phone sign-ons are all powered by deep learning,” Puthawala said.

Puthawala and Furuya met in 2021, when they were both postdocs. Puthawala was at Rice University, and Furuya made a visit there.

Trip identifies future potential collaborators

They hit it off relatively quickly and have since published two papers on deep learning with more in the works, Puthawala said. Given their shared interest and existing relationship, they were able to submit a solid proposal to the NSF. They learned of their selection in in time for Puthawala xto adjust his teaching load in the fall to accommodate for not teaching spring semester.

Part of his time in Japan was spent in Furuya’s office processing math problems. He also attended two math conferences in Kyoto and another in Kobe, where he also visited a faculty member.

Puthawala said his branch of mathematics doesn’t have its own organized society, but there is an informal network of people who study this specialized area.

“Ninety percent people using deep learning are using it to solve a problem,” such as fraud detection or facial recognition. “I don’t do anything useful,” Puthawala said. Further explaining, he said his work is basic science. “Right from the beginning, I wanted to focus on the fundamentals of deep learning.”

Deep learning work dates to 2019

Puthawala’s start in this branch of mathematics dates back to 2019, when he began postdoctoral work at Rice.

“My adviser suggested I do machine learning, which at the time was novel. I felt I needed to distinguish myself, to develop a niche.” That niche was studying the fundamentals of machine learning. “To make advancements, you don’t need to run a lot of code. You just have to think about it in a way no one else has,” he said.

He provided this analogy. “To be an Olympian, you not only need to be a great athlete, you also need to have access to great coaching, great facilities, the right training partners and proper nutrition. But to be a great writer, you don’t need a lot of infrastructure. It’s the same way for those who study the fundamentals of deep learning.”

Puthawala said he doesn’t use the university’s high performance computing lab. He just uses his desktop computers.

No need for high-performance computers

“I’m not using computers to run a science model.  It’s a lot of chin rubbing and calculations that I could do with paper and pencil if I were more patient.”

Puthawala said that in rubbing shoulders with his Japanese counterparts, he gained information that will someday turn into a paper. “It was super useful for personal research, and I met with some of Fururya’s students, who expressed an interest in a summer aboard study,” he said.

Puthawala added that he will continue to nurture the new faculty collaborations he made and hopes someday for a reciprocal trip.

He adds that he appreciates the hospitality shown by Furuya, who helped him make connections and acclimate him to life in a foreign country.