Valerie Chen (MY ’20) is no ordinary computer scientist. Motivated by the applications and impact of technology, as well as the difficult questions surrounding building intelligent systems, Chen has conducted research on these fronts, collecting prestigious awards from the White House, Qualcomm, Snapchat, Microsoft and more along the way.
Growing up in Northern Virginia, Valerie attended a technology-focused high school, taking STEM classes—including computer science—as part of the curriculum. However, an internship at the Naval Research Laboratory in DC first truly ignited her passion for computer science. “The experience changed my perspective on technology—the work I did made me realize the real-world applications and effects that technology can have on the broader scale,” Chen said.
The Naval Research Laboratory internship ignited Chen’s passion for research, but she had also explored software engineering briefly. While she found it interesting to work at a large tech company, she “feels like there is a lesser focus on creativity and more focus on the ability to get the program to run in the long term.” Chen prefers being able to think outside of the box without being limited by the market value of her work. She adds that computer science research is unique in its incorporation of math, psychology, social interaction, and many other disciplines. But most important for Chen is how tangible computer science is. “If you want people to test something you designed, that can easily happen and implementation is generally fast,” Chen explained.
Despite involvement in a multitude of research projects, it was only recently that Chen started feeling comfortable going through the process of creating a hypothesis and conducting her own research. “Starting out in any research field, there is a steep learning curve. You need to not only understand what current methods and ideas are, but also what researchers have already come up with in the past,” Chen said. It was only last year when she felt capable of coming up with her own project in Professor Zhong Shao’s lab.
Shao, a professor of Computer Science at Yale, wanted to implement a more secure operating system in self-driving cars. Chen proposed a novel approach to use these self-driving cars as a platform to collect and build a machine learning model from the autonomous driving data. Rather than seeing if she could improve the precision of the self-driving car, she suggested that there could be ways to identify patterns in the car’s faults given a pre-trained decision-making model. For Chen, novelty and fault detection are crucial in safety-critical deep learning models, like those behind self-driving cars, because of the lack of guarantees in current machine learning algorithms. “There are some fundamental basics of human decision-making that we cannot quite imitate with robotics yet. My hope was to be able to pinpoint those discrepancies by systematically learning a representation of task-aware features to identify situations where the machine will make an incorrect, untrustworthy prediction,” Chen said. In the end, her project was successful. Her methods outperform prior work on identifying novel scenarios—from recognition of different driving environments to different inputs changed through adversarial attacks.
Most recently, Chen has been working on a project in collaboration with a lab in the Robotics Institute of Carnegie Mellon University. With their help, she is developing efficient methods for solving hierarchical problems—problems in which structural knowledge and low-level skills are needed to accomplish high-level tasks—through imitation and reinforcement learning. This work is relevant for robots who in the future will have to reason and make decisions about complicated tasks. Through this project, Chen dove into the one area she finds most interesting: the intersection of machine learning, robotics, and psychology. “I find it interesting that even psychologists don’t know exactly how the human brain works—this leaves computer scientists to think about approaching psychology from a computational sense,” Chen said. Chen is one of the many individuals working towards answering that question, but, in Chen’s own words, “research is incremental.” She finds immense value in working on small projects that build upon the work of others to reach that goal.
In terms of advice for aspiring researchers, Chen emphasizes passion. “Research is full of failed experiments and difficult implementations—it is important to find that drive and passion to push you past a rough patch.” Chen’s passion and resilience will not only continue to drive her to better the world, but inspire those around her as well.