When Chen Xinyue '22 arrived as a first-year student at NYU Shanghai in the fall of 2018, she had never before written a line of code or even taken a class in computer science. Still, when choosing her first-year classes, she decided to give Introduction to Computer Programming a shot.
"I knew there are stereotypes that dissuade girls from approaching computer science, but when I talked with academic advisors about walking into a computer science course with no prior programming skills, they were quite encouraging and gave me many examples of other female NYU Shanghai students who had excelled in the field," she says.
Now, four years and two published papers later, Chen is weighing full scholarship offers from computer science Ph.D. programs at some of the world's top universities, including the University of California at Berkeley, the University of Toronto, Cornell University, and the University of Texas at Austin.
Not long into her first semester, Chen, a Shanghai native, discovered that she and coding were a perfect match. She enjoyed using code to express her ideas - and was very good at it. "I realized that logical thinking is more important than the coding skill itself, as the latter is just a tool to implement the idea."
After excelling in Introduction to Computer Programming, Chen was inspired to be even more ambitious for her next semester. She enrolled in Dean of Engineering and Computer Science Keith Ross' course, Machine Learning, becoming the only first-year student in a class full of upper-class students.
"Machine learning is the foundation of artificial intelligence which is very hot right now and is about improving the performance of specific algorithms in empirical learning," Chen says. "After taking a basic computer course, I wanted to…challenge myself."
Chen was impressed with Ross's approach to teaching. "People might think that in Machine Learning, all you need to do is learn how to write code and do data analysis, but Professor Ross will spend nearly half of the class time explaining the underlying theory and mathematics knowledge behind it, which gives us a very solid foundation," she says.
Chen's confidence as a student also blossomed in an engaging learning environment. After explaining a theory, Ross would often turn to the class and ask questions. "I would just speak up whenever I had an idea, without worrying if I was wrong," she says. "Professor Ross is always encouraging us to share our thoughts. Sometimes, even if our answers were just partially correct, we would still get praise from him, which gave me a lot of confidence."
In the Windows Lab of the Academic Building during her freshman year
Since the summer after her freshman year, Chen has been a member of Ross' machine learning research team of two Ph.D. students and three other undergraduate students. With the support of the Deans' Undergraduate Research Fund (DURF), Chen has explored the use of deep reinforcement learning in training robots in simulation environments to perform various tasks such as running.
One important lesson Chen says she learned from Ross and his team was the value of being rigorous. "I remember once, my algorithmic experimentations led me to what seemed like promising results, and I reported back to Professor Ross. But he asked if this was 'an average result obtained by carrying out several runs of experiments and was robust enough to reach a high performance,'" she says. "You can never be too careful in research. If you feel that there is hope in a certain direction, you should verify that conclusion with many experiments."
Her hard work in research has paid off. With Chen as the first author, Ross' team published a research paper BAIL: Best-Action Imitation Learning for Batch Deep Reinforcement Learning, which was accepted by Neural Information Processing Systems (NeurIPS), a top international conference on artificial intelligence. That first paper has been followed by two others. The paper Randomized Ensembled Double Q-Learning: Learning Fast Without a Model, published by the International Conference on Learning Representations (ICLR), one of the top conferences on machine learning in the world, remains one of her proudest achievements.
"Xinyue has outstanding knowledge and skills in programming, mathematics, and algorithms. She is already an expert on deep learning and on reinforcement learning," Ross says. "Although she is an excellent student with a very high GPA, her greatest strength is in creative research. She has both a passion and gift for research, and already has the research maturity of a seasoned Ph.D. student. It has been a true pleasure working with her the past few years."
While Chen is still weighing her graduate options, she is certain of her research interests. "I hope to do new research in machine learning in the future and to become an expert and push the boundaries of the field. Even if it's just a little bit of progress, it's still my own contribution to human society," Chen says, "My belief is that machine learning research should be valued both by its experimental performance and its potential impact on shaping our world. I always pay close attention to the fairness issue in machine learning research. I believe a significant reason for the 'unfairness' of machine learning models is their unbalanced datasets, and I aim to make progress on this issue whenever possible. "
Looking back, Chen says she has been most grateful for the chance to explore her interests freely and flexibly. She says it is unlikely she would have discovered her passion for computer science if she had not come to NYU Shanghai and instead gone to a Chinese university where she would have had to choose another major based on her score on China's national college entrance exam.
"At NYU Shanghai, we are encouraged and inspired by others to find our true interests and potential - leapfrogging into progress instead of setting limits," she says. "I have encouraged and advised younger students to get involved in computer science research. Many of them are women. I tell them, 'Don't be afraid to try things you haven't tried before, and don't feel that you must be prepared for everything before you can take up a subject. Only by taking the first step can you find out if you're suited or not. If it's something you're really good at, you'll be fully motivated to break the limit and keep exploring and widening the boundaries.'"