Yuhan Pan is a first-year graduate student who has been enrolled for nearly a year. During the three-month freshman summer camp, he successfully submitted a paper and was selected as one of the “Outstanding Campers” for his excellent performance. We invited Yuhan Pan and his class monitor Zhipeng Zhou during the summer camp to participate in our interview, discussing how he gradually overcame difficulties and adapted to graduate studies from a novice in scientific research.
Chenyang Lin: Yuhan Pan, congratulations on being selected as an excellent camper of the Freshman Summer Camp. Can you briefly introduce your summer camp project to us? And what is the reason for choosing this topic?
Yuhan Pan: The goal of our project is to enhance the performance of existing WiFi based human behavior recognition models through adversarial training. Due to the disturbance of CSI data used in HAR, which is not as intuitive and visible as image or text data, the quality of adversarial samples becomes difficult to control, leading to poor results in adversarial training. Based on this issue, we designed methods such as confidence calibration and sample selection to only retain the high-quality portion of the generated adversarial samples, thus ensuring the effectiveness of adversarial training. The final experimental results indicate that our method can improve the accuracy and robustness of existing HAR models. This is the goal and proposed method of my summer camp project. When it comes to choosing this topic, it is actually because at the beginning of the summer camp, four class monitors came to give lectures. I am not familiar with the topics taught by other senior students, such as Backcatcher, because I do not have a foundation in this field in my undergraduate studies and am worried that my foundation is relatively weak and my abilities are insufficient. Although my foundation in the field of deep learning is not very solid, after all, I worked on the corresponding projects during my graduation and it doesn't look so unfamiliar, so I ultimately chose the direction of deep learning.
Chenyang Lin: During your summer camp, you successfully submitted a paper. During this process, what difficulties did you encounter that left a deep impression on you? How was your mood at that time?
Yuhan Pan: There are actually quite a few difficulties, after all, it was my first time being exposed to a relatively formal scientific research project. I think there are three main difficulties. Firstly, there were difficulties in reading the paper. I remember that Zhipeng Zhou sent us a few references and when I saw all the English literature, my head was spinning and I felt a little overwhelmed. Reading was slow. For some of the more esoteric mathematical reasoning, I sometimes dug in, rather than grasping a paper as a whole, and did not have a good habit and method of reading papers. Secondly, there are difficulties in understanding the code. Senior brother provided us with some reference code. I remember reading it for several weeks, and I often have some areas that I don't quite understand. I consulted senior brother at the weekly class meeting. The third point is the difficulty in coding. I remember at the beginning, I even forgot some basic Python syntax and lacked knowledge about deep learning frameworks. For example, the Python framework, the splicing, dimensionality reduction, and dimensionality enhancement of Tensor data have all troubled me for a long time. Faced with these problems, I sometimes have an unstable mindset and exclaim, Wow, it's so difficult. Sometimes, I also feel restless and wonder if scientific research is really suitable for me. Then, I may need to calm down and adjust my mindset to continue doing it.
Chenyang Lin: Your summer camp was not smooth sailing either. How did you overcome these difficulties? Can you provide us with a specific introduction to your method?
Yuhan Pan: First of all, when my mentality is not very stable, I will calm myself down for a moment and not focus on this topic. Because it was summer vacation at that time, I might temporarily put this topic aside when my mindset was unstable and start doing something else to adjust my mindset, such as playing games or going for a walk. Because I know that when I am very upset, I cannot continue with the project. Secondly, for specific technical issues, I will search for relevant materials online and watch related videos to supplement and learn some basic knowledge, so that I can continue to work on this topic. Of course, another important point is that at that time, Zhipeng Zhou held a small meeting for us every Saturday. There were some questions that could be discussed with him during the meeting, and he would provide us with certain answers. Although I have encountered many difficulties overall, through the methods mentioned earlier, I have successfully overcome these difficulties and completed the summer camp project.
Chenyang Lin: I think some of the experiences you shared are very practical. Speaking of Zhipeng Zhou's help, we also specifically invited him for this interview. Zhipeng Zhou, what kind of mentality did you have when guiding your younger siblings? Have you seen yourself when you first started researching in them, and did you also receive guidance and assistance from your teacher or laboratory colleagues at that time?
Zhipeng Zhou: Firstly, throughout the summer camp, students such as Yuhan Pan, Luyao Liu, and others were very positive themselves. This is especially true for me and most of the new students who just entered the laboratory. This positive energy can last for at least one semester. Secondly, I felt that having a senior student with a new student in hand can really save a lot of time. I used to be in a laboratory at the Information Technology Institute, and as soon as I joined the team, I was the team leader of the project, and I was working alone, which is similar to Yuhan Pan's current state of working on the bank project. I also do some research on my own besides working on projects, because my supervisor is busy, and my senior brothers and sisters are also busy looking for jobs or applying for overseas PhD. Of course, whenever I have any questions to ask them, they are still willing to answer me, but most of the time I can only explore alone. At the beginning, I was not familiar with anything. It's really difficult to rely on my own exploration, because I need to go through all the problems from beginning to end, and I may even get stuck in the initial stage of configuring the environment. At that time, Cafe and Tensorflow were popular, and Cafe needed to be compiled from scratch. It took me about several weeks to reluctantly prepare the environment. However, if someone took me along at that time, this problem might have been solved in two days. So I sincerely hope that I can provide some of my own experiences to the new students as much as possible, which can be considered to make up for my regret to some extent. Additionally, I think the atmosphere of mutual assistance in the laboratory I used to work in and our current one is particularly good.
Chenyang Lin: In the future, Yuhan Pan will also become a senior student. He may guide and help new students. Do you have any thoughts on this?
Yuhan Pan: I am very lucky to have a summer camp experience after joining the group. The help of my teachers and senior students saved me many detours. So, if in the future I have the opportunity to change my identity and become a group leader to guide and assist freshmen, I believe this will be a very good opportunity to pass on my accumulated experience to freshmen. Specifically, I feel that Zhipeng Zhou's guidance during the summer camp is suitable. He holds meetings with us every Saturday to discuss the problems encountered this week first, and after solving the problems, he will also make a plan for the next cycle. I think this is a good approach because it can help students who are new to scientific research have a clearer understanding of what they have done and what they will do in the future, so as not to feel confused and aimless alone.
