Recently, the National Natural Science Foundation of China (NSFC) announced the review results of the 2025 National Natural Science Fund for Young Student Basic Research Projects. After a highly competitive selection process, Tian Yongming, an undergraduate from the Class of 2022 in our school, successfully secured the grant.Under the guidance of his mentor, he has been delving deeply into the field of deep learning, continuously exploring its core mysteries. Using the "Deep Neural Kernel" as a key, he attempts to unlock AI's black box. Today, through an in-depth interview, we explore the thought processes behind his scientific research journey.
Q：Where did the inspiration for your project originate, and how did the initial concept take shape into an executable plan?
A：The inspiration for this project is rooted in my long-standing research interest in the field of deep learning and sprouted from the knowledge and experience accumulated through past studies and practice.On one hand, I have been deeply impressed by the powerful representational capacity and generalization performance of deep learning, which also sparked a strong curiosity about its underlying principles. Traditional machine learning theory suggests that more model parameters tend to lead to overfitting, resulting in a decline in generalization performance. However, in deep learning, the "double descent" phenomenon exists—where test error exhibits a pattern of "first decreasing, then increasing, and finally decreasing again" as the number of model parameters increases. The "final descent" phase, in particular, highlights the exceptional generalization ability of deep learning. So far, however, existing theories have yet to provide a comprehensive and thorough explanation for the logic behind this phenomenon. Deep learning remains, in many ways, a black box. This gap between its remarkable capabilities and the unknown nature of its internal mechanisms has fueled my desire to explore it in depth.On the other hand, with the encouragement and guidance of Professor Zhang Shaoqun, I conducted extensive literature reviews and preliminary experiments, learning that the deep neural kernel method could be an effective tool for analyzing the representation and generalization capabilities of deep learning. Therefore, I decided to apply for this project, hoping to uncover the mechanisms behind deep learning from the perspective of neural kernel learning and provide a reliable theoretical foundation for the design of next-generation deep learning algorithms.
Q：Reflecting on your journey of applying for the National Natural Science Foundation of China (NSFC) grant, what was the single biggest challenge you faced, and how did you navigate through it step by step to reach where you are today?
A：During the application process, I faced difficulties in two main areas. The first challenge was time management. The grant application was extremely time-consuming, and the application period overlapped with the summer camp and pre-exemption process for graduate recommendation. During that time, I had to juggle multiple tasks simultaneously, which tested not only my time management and energy allocation skills but, more critically, my ability to adjust my mindset promptly. To balance all these commitments, I tried to maximize my time efficiency, and when I was truly overwhelmed, I had to work late nights to meet deadlines.The second, and most challenging, difficulty was writing the proposal. I had no prior experience in writing grant applications, and transforming a good research idea into a logically clear and well-structured proposal was indeed very difficult. To break through this bottleneck, I had no choice but to revise and polish it repeatedly, enhancing its coherence and logical flow through continuous refinement. During this iterative process, my advisor, Assistant Professor Zhang Shaoqun, provided immense help. He not only pointed out the immature parts of my thinking but also taught me how to better present research ideas and standardize language expression. Professor Zhang's guidance helped me avoid many detours and gradually understand the logical framework and writing standards required for a high-quality proposal. Under his careful supervision and through repeated revisions, I ultimately broke through the bottleneck and truly mastered the essentials of writing a successful application.
Q：What specific support did your mentor, your school, and the university provide during your application for this project?
A：During the application process, I received comprehensive support from multiple sources.First and foremost, I received the full support and guidance of my supervisor, Assistant Professor Zhang Shaoqun. He not only guided me on how to transform a research idea into a high-quality proposal but also provided invaluable advice on writing and defense preparation, which helped me avoid many potential pitfalls. Furthermore, his consistent encouragement gave me the confidence to persevere and complete the entire application.At the school level, I was provided with essential resources and a conducive working environment. This support enabled me to smoothly conduct preliminary research in the early stages of the project and to prepare application materials efficiently later on. Additionally, the school organized research training sessions and experience-sharing activities, from which I gained many valuable insights.Lastly, at the university level, Nanjing University has established a well-structured project application system, allowing me to clearly understand the requirements at each stage. More importantly, the university organized multiple online and offline training sessions for all applicants. The instructors in these sessions were highly dedicated and offered numerous constructive suggestions. These trainings also provided a platform for us to learn from each other's application approaches and presentation methods, identifying and adopting each other's strengths through mutual exchange. This atmosphere of collective learning and shared improvement is, in itself, a precious asset provided by the university, offering the most solid support for our growth.In summary, the guidance from my mentor, the nurturing environment of my school, and the robust support from the university gave me the foundation and confidence to successfully complete this application.
Q：What are your key strategies for handling the dual pressures of research projects and coursework?
A：Both research projects and academic tasks require a significant amount of time. My approach is to try to make them complement each other and promote each other. For instance, when I come across theoretical knowledge in my course study, I will attempt to apply it to my research projects to deepen my understanding. Conversely, the problem-solving skills and thinking patterns cultivated during the research process will also enhance my academic performance. Additionally, I believe that the time allocation for studies and research can be adjusted. In the early years of university, the focus should be on laying a solid academic foundation. As one progresses to the later years, more time can be devoted to research, applying the knowledge acquired to solve practical problems.
Q：Among the many popular research directions in artificial intelligence, how do you view "chasing hotspots" and "engaging in valuable niche research"?
A：I prefer to understand this issue as how to strike a balance between "chasing hotspots" and "engaging in niche research", because there is no necessary connection between the "popularity" of research and its own "academic value", which requires a dialectical view. On the one hand, hot research topics usually attract a large number of scholars to engage. If the implementation difficulty of a certain topic is too low, the value of the final outcome is likely to be discounted. It may even lose its significance because it has been preemptively accomplished by others. On the other hand, unpopular research does not mean it is "valueless". Professor Zhou Zhihua has always encouraged everyone to venture into the "uncharted territory" of scientific research, precisely because many niche directions actually hold great potential, but they just temporarily lack appropriate research methods. It is precisely because of their obscurity that once these directions are broken through, they often hold the key potential to drive the development of the entire field. Meanwhile, the popularity of research also changes over time. What is popular may cool down, and what is unpopular may become popular. What we can do is to grasp the intrinsic value of research work, which is the most fundamental pursuit.
Q：Do you have any suggestions for your younger schoolmates in terms of research and studies?
A：In terms of scientific research, I am just getting started myself and still need to keep accumulating and refining. Based on my experience, I mainly have two suggestions. First, communicate more with classmates, senior students and teachers. When people first get involved in scientific research, most of them do not have a clear understanding of it. Moreover, scientific research is actually quite different from the academic studies we are familiar with. Therefore, it is essential to maintain a humble attitude, proactively seek advice from experienced people, and establish a correct understanding of scientific research. Secondly, scientific research is often described as "smart people doing foolish things", but in fact, this statement means that scientific research requires long-term and repetitive investment. We have to spend a great deal of time repeatedly reading literature, patiently conducting experiments and constantly revising papers. Many times, we don't see obvious results in the short term. But it is precisely these seemingly "clumsy" persistences that are the key for us to accumulate experience, cultivate scientific research thinking and capabilities. Only through such repetitive attempts can one gradually grasp the essence of scientific research and form one's own understanding and methodology. In terms of studies, I always believe that no pains, no gains. I suggest that everyone plan their daily study time reasonably. At the same time, attention should also be paid to methods and efficiency. Personally speaking, I am more efficient in the morning and it is suitable for learning new knowledge. It is more suitable to consolidate what has been learned by doing exercises in the evening. No matter when you study, concentration is the key. It's best to mute your phone to ensure you are fully focused.
In this conversation, we witnessed Tian Yongming's sincere love for scientific research and his in-depth thinking. From confusion to exploration, from trial to breakthrough, his experience vividly demonstrates the perseverance and growth of Nanjing University students on the path of scientific research. May he always maintain this enthusiasm for exploration in his future academic journey and go further and more steadily on the road of AI theoretical research.
