ESE PhD candidate Behrad Moniri recently defended his thesis on Understanding Deep Learning Via Tractable Models. Advised by Dr. Hamed Hassani, Moniri’s research focuses on uncovering why heuristic methods play such a large role in the empirical success of deep learning systems.
According to Moniri, the empirical performance of deep learning systems is “largely driven by heuristic methods.” That said, “We do not really know why do they work so well. My research focuses on building the mathematical foundations of deep learning and trying to better understand the reason for their success,” Moniri explains.
“By developing and analyzing rigorous mathematical models, I aim to better understand these systems and use this understanding to improve their performance, efficiency, and safety,” Moniri adds.
Moniri’s success as a researcher did not come without its challenges. Throughout his time at Penn, Moniri grappled with maintaining work-life balance. “I can’t say with a straight face that I have been able to overcome this challenge! I had to set specific time limits on work to enforce the balance to some extent,” Moniri explains.
Challenges aside, Moniri considers himself “very fortunate to be able to collaborate with my brilliant mentors and co-authors. Working on these challenging and fundamental questions with these extremely smart people has been my most memorable memory,” Monir says, adding, “We had many stimulating conversations that shaped the way I think about the world!”
The collaborators and mentors mentioned above all played a pivotal role in Moniri’s success. “I would like to thank my advisor, Prof. Hamed Hassani, for his amazing mentorship,” Moniri emphasizes. “Also, I like to thank my collaborators Prof. Edagr Dobriban, Dr. Donghwan Lee, and Dr. Thomas Zhang.”
Further, Moniri extends deep appreciation to his family and friends. “I would have given up long ago if it wasn’t for their constant support and encouragement.”
In addition, Moniri extends gratitude “to all of the ESE department staff for their help throughout my years at Penn.”
When it comes to future goals, “I plan to continue working on the same research program to try to better understand deep learning systems,” Moniri explains. “To do this, however, I think it is necessary to gain more hands-on experience actually building such systems,” Moniri explains. “This is currently impossible in an academic setting because of resource constraints. As a result, I plan to work as a machine learning researcher in the industry for a few years to gain this experience. “
When Moniri’s not working, he enjoys watching movies, reading history books, and following current events. “I also love to travel,” Moniri adds.
Learn more about Moniri’s work here