The journey of pursuing a Ph.D. is an incredible experience. Collaborating with talented individuals and tackling meaningful challenges is deeply rewarding. However, as I reflect on my journey, I recognize several mistakes I made along the way. I’ve decided to share these insights in the hope that others can benefit from my experiences. I believe these lessons extend beyond just the Ph.D. process and are applicable to many other long-term and challenging endeavors.

Too Divergent and Lacking Focus

Honestly, I’m a divergent person, full of curiosity for various fascinating things and eager to explore them all. During my Ph.D. journey, I worked on numerous diverse tasks, including “Human/Hand Mocap,” “NeRF & Reconstruction,” “Unsupervised Representation,” “Motion Generation,” and “Image/Video/3D Generation.” While trying different projects gave me a broad understanding of AI, it also caused me to lose focus.

I could argue that my overarching goal was pursuing general intelligence to understand the 3D real world. Many researchers use similar high-level goals to summarize their work. However, in my view, this approach is not ideal for a Ph.D. student. With limited time, an AI Ph.D. student should concentrate on a small but meaningful sub-goal within AI.

So, be focused. Generally, there are two types of focus. One is task-oriented, such as working on image detection throughout the entire Ph.D. journey. The other is goal-oriented, which I call “goal-driven research.” The goal must be technically challenging or practically useful. For example, you might define your goal as building a capable Mocap system. Then, you would identify the challenges and tasks needed to achieve this goal, such as solving 2D human detection, 2D human pose estimation, and 2D-3D human pose lifting. Always focus on the most valuable and overlooked aspects of the process. In a goal-driven Ph.D., you act as a problem solver, selecting or creating the right technologies to tackle the challenges.

Stay focused on your mission and goal. By the time you graduate, you may have produced a series of works that address significant problems, adding new value to the community. Your contributions will be undeniable, forming a strong foundation that showcases your value.

Left: A good reminder from Bill Freeman; Right: Slides from Jason Wei.

Throughout the process, always remember that one outstanding paper has far more impact than numerous average ones. Many Ph.D. students understand this in theory, but it’s challenging to adhere to in practice. I found myself involved in numerous projects, often feeling drained by mundane papers and tasks. Honestly, a groundbreaking paper or an exciting project brings me significantly more joy and satisfaction than a routine one.

Instantiate Intermediate Results

Another mistake I made was overlooking the importance of “instantiating” intermediate work. By “instantiating,” I mean documenting your explorations and writing down your ideas and understandings throughout the process. This could involve writing blogs to share your thoughts, creating GitHub repositories to summarize your work and provide useful tools for the community, or making project demos and posting them on Twitter. Always remember to instantiate your work, making it something that others can understand and use.

Why should we do this? There are two main reasons. First, it’s an effective way to showcase your knowledge and ideas to others. While papers are a classic form of instantiation, they’re sometimes not enough. Blogs, Twitter, and other platforms are also valuable mediums for representation.

Second, it helps you build a feedback loop for yourself. Learning is a spiral process: initially, you know almost nothing about a new domain. After some time, you feel like you know a lot, but as you learn more, you realize how much you still don’t know. Documenting your intermediate steps helps solidify your understanding and reveals gaps in your knowledge, allowing for continuous improvement.