Hi, I am Yuxi Zhu, a Bioinformatics and Computational Biology specialist and Molecular Genetics Major who just finished my second year. Like most people, this is my first formal research experience. Professor Tyrrell warned me from the start that I would need to be independent in this lab, but my genuine interest in ML and its applications gave me the confidence to take on the challenge. Overall, this summer’s ROP journey in the MiDATA lab was filled with both excitement and challenges.
The first challenge was finding a research question. I’m incredibly grateful to Daniel, a volunteer and former ROP student, who introduced me to the concept of “adversarial examples” and helped me formulate my research question from the start. During the first two months of the literature review, I often found myself diving too deeply into theoretical aspects that were less applicable to Medical Imaging, or exploring questions that, while feasible, didn’t capture my interest. Luckily, I was able to settle down with understanding the differential effects between random perturbations (like random noise and loss of resolution) and non-random adversarial perturbations on the model.
As the project progressed, I encountered a series of obstacles and bugs that required constant problem-solving and debugging. For example, my initial findings showed very low performance, all under 50%. Professor Tyrrell pointed out that the accuracy of a binary classifier should never drop below 50%, as that would mean it’s performing worse than a random model. I quickly realized there were bugs in my code and implementation. Additionally, after obtaining results, I thought interpreting them would be straightforward. However, when Professor Tyrrell asked me why adversarial perturbations led to accuracies below 50% while the others didn’t, I found myself at a loss for words. In the end, with Professor Tyrrell’s guidance, I was able to interpret the results correctly and articulate them in my report.
Despite the stress I felt before presenting my findings at our weekly meetings, these sessions became invaluable learning experiences. Professor Tyrrell would scrutinize my work with questions and critiques, pushing me to think more deeply and critically about every aspect of my research. The other lab members also provided very helpful insights and shared their work. These meetings not only allowed me to understand what others were working on but also gave me the chance to get involved in or observe lively discussions that often took place.
Looking back on the last few months, this experience has been invaluable. I am deeply thankful to Professor Tyrrell who offered me this wonderful opportunity in ML and guided me through my research project. I especially appreciate how we weren’t just taught to implement a given research project or conduct a specific experiment; we were taught how to find gaps and how to conduct research. I also want to express my gratitude to Daniel for his support and insights when I was in doubt, and to Atsuhiro for his helpful suggestions. Completing my first-ever research project was challenging yet rewarding, and I am grateful for all the guidance and help I received. I’m confident that what I have learned will stay with me in my future research and career.