Yuandong Tian:A Five-Year Reflection on My PhD Experience
Yuandong Tian: A Five-Year Reflection on My PhD Experience
Over the past five years, the most important thing I’ve learned is how to approach a substantial piece of work. It sounds lofty, but it really comes down to one point: you must quiet your mind.
First, only with a calm mind can you immerse yourself deeply enough in a field to do serious work. Today there is an overwhelming amount of social media and endless streams of news. Spending every day responding to these broad yet shallow signals, or busying yourself commenting on others, makes it impossible to get anything meaningful done. It’s like traveling: following a guidebook from one famous attraction to another brings no real insight—it only adds a bit of conversational material. To truly appreciate the beauty of nature, you must go where others cannot reach; you need goals, patience, perseverance, and long-term planning.
Second, even with determination and willpower, you can’t succeed without method. Every step of research requires methodology: how to conduct basic background work, how to get started, how to analyze problems, how to defend your viewpoint, how to prioritize tasks, how to handle the relationship between details and the big picture, how to gradually turn intuition into rigorous mathematical language and verify it. Getting stuck at any point can halt your progress. And since no one can be fully prepared in all these aspects in advance, the only feasible path is to set a direction and then explore step by step.
At the beginning, you will inevitably do a lot of useless work: literature searches will feel directionless, you won’t know how to choose a topic, you lack experience so you cannot be efficient, and inefficiency breeds impatience, which leads to giving up and denying everything you’ve already achieved. Only by forcing yourself to sit down and systematically analyze past mistakes can you break this vicious cycle. Gradually, experience accumulates; you start seeing paths you couldn’t see before. To reach this point, a calm mind is essential.
Third, research inevitably comes with difficulties: an unhelpful advisor, uncooperative data, formulas that refuse to work out, a computer that breaks, and so on. During the PhD, you could come up with a thousand excuses to blame others, but I believe the best attitude is “no complaints, no explanations.” Do what is within your power, take responsibility for your mistakes, bring them home and analyze them—only then can you improve. You must have ideals and convictions; otherwise how do you keep going in this lonely marathon? But don’t set the tone too high—an unrealized ideal is worthless to others. Achieving all this also requires quiet perseverance; no one cheers for you when you overcome difficulties.
The essence of research is free exploration under a relaxed environment, where researchers seek answers to the unknown. No matter how great an advisor is, they only have vague intuitions; unless they participate in every detail of the research, they cannot know the precise answer. Precise answers must come from the student; errors in the advisor’s intuition must be overturned by the student. Without inner motivation, the advisor can only see what they themselves are capable of seeing—there would be no need for the student, and the research would fail.
Thus, the success of research ultimately depends on the researcher’s inner drive for unceasing exploration. If you see yourself as someone who prefers comfort, follows orders, and doesn’t like to think, then a PhD may not be for you. If pressure from family or peers outweighs your own desire, the PhD may become a personal hell. And if your only goal is to learn more specialized knowledge, the enormous pressure of the dissertation in the later years may make a master’s degree a more sensible choice.
Fourth, read more of others’ work—but not too much. Grasp the main thread.
The number of papers published each year has been rising; there are simply too many. Reading each one carefully is a waste of time and yields little. In my opinion, the best approach is this: after reading and summarizing a few of the most important papers in the field, try to infer the general strategies people use and the strengths and weaknesses of their approaches. Then, when reading new papers, use a skipping-and-guessing method—read selectively and predict as you go. If you guess right, you reward yourself. This method is not only fast but also gives you a grasp of the big picture while filtering out trivialities. When developing new ideas, you naturally avoid common paths and aim for the vulnerabilities of existing work, ensuring the novelty of your own. This may seem contrary to the ideal of academic rigor—but life is short; do what works.
Returning to a previous topic: when I first entered CMU, I didn’t manage to join my ideal advisor’s group for various reasons. I felt at a loss and asked a senior for advice. He told me, “This isn’t what matters. What matters is what you can learn from him. I think he is a good person, and his presentation and writing skills are excellent.” As it turned out, he was right.
When choosing an advisor, the specific research topic is not the most important factor. More important are their character, communication, and expression skills. My advisor, though Indian, has excellent character—strict yet responsible, and he never delays students’ graduation. What I learned most from him were two things: presentation and writing.
For students, identifying the main line and discarding details is painful. Students often miss the forest for the trees, unable to distinguish what truly matters. Instinctively, they treat the parts they spent the most time on as the “key content,” and when preparing a talk, they become reluctant to let these go. Or they feel overwhelmed, unable to decide where to begin. Should you explain the details of the objective function? The gradient descent formula? The data normalization? A small trick that improves performance by five percent?
The answer always depends on the goal. If the work is about improving performance systematically, then small tricks become the main storyline. If it's a new algorithm, then the clever construction of the objective function is the centerpiece. If the research focuses on data statistics, normalization becomes crucial. If discussing large-scale distributed feasibility, then dependencies among parameters and computational complexity are key.
In short, if a detail does not contribute to the grand structure you are building, then it does not deserve lengthy discussion in a presentation. By contrast, advisors—precisely because they are not immersed in every detail—can more easily see the big picture and understand the true importance and interconnections of components. They can make decisive trade-offs.
A good slide can take many forms: a single large figure, several interconnected images, a list of prior works, a pros-and-cons table, the three main steps of an algorithm, or diagrams showing relationships among concepts. In short, if someone stares at it for ten seconds and still cannot tell what the main point is, send it back and redo it. Making beautiful slides is an art; I’m still not even one-tenth as good as my advisor. So I keep looking at good examples and improving slowly.
In addition, exercising is essential—you must not neglect it, no matter how busy you are. Spending an hour or two exercising relaxes the mind, rests the brain, improves cardiovascular function, strengthens immunity—it's a guaranteed win. I usually run five kilometers every other day, or swim one kilometer, or play squash/badminton for an hour. Running too frequently can hurt the knees, so it must be combined with other activities. While running, I think about nothing. In those sweaty moments, I am simply happy.
As a CMU classmate once said, there is no such thing as a “lifetime iron rice bowl” anymore. Companies operate for profit; good salary and benefits today may vanish tomorrow if performance drops. Being laid off five or ten years later—when you have aging parents and children to support—means falling from heaven to hell. Therefore, crisis awareness and long-term planning are crucial. As we grow older, our corresponding capabilities must also mature. The goal is to broaden our future choices. Thus, we should invest our time in skills that age well: communication and writing, systematic knowledge, practical problem-solving experience, deep understanding of fundamental tools and principles. With these, no matter what role you take in the future, you can quickly integrate into a team, understand others’ needs, grasp problems deeply, and produce results quickly.
First, proceed step by step. Envying others’ quick success is pointless—you see only the glamorous surface, not the hard work underneath. Almost any job has hidden potential; if you feel genuine enjoyment in thirty to fifty percent of your work time, that’s already very good—no need to switch. If you truly dislike your current job and want to switch, the prerequisite is that you must first do your current job well. Only then will you have the financial foundation to change, enough justification to change, and the mental space to do so.
Changing fields isn’t done on a whim. It happens gradually: first interest, then hobby, then side project, and eventually a mature skill, which becomes the main career. At each stage, if setbacks occur—or if family emergencies arise—you can always retreat safely, maintaining life stability. During my PhD, I followed this principle. Even though my advisor’s topic wasn’t exactly my preferred direction, I didn’t switch groups; instead, I completed my advisor’s tasks while slowly building my own direction and eventually turning things around.
Thus, when entering a field you’ve never tried before, you must be cautious. You must constantly evaluate whether your thinking style is suitable, and whether you can acquire the necessary sensitivity within a limited time. Everyone has a type of work they are naturally suited for. Progressing along the thinking mode that fits you best is the most stable, practical, and efficient approach.
Someone might ask: what if you're starting from a blank slate, with no experience and no established way of thinking? In that case, you can only rely on time and patience to build things up, while frequently seeking advice from seniors. After two years in a field, should you persist or give up? No one can say. Sometimes persistence leads to breakthrough; other times it only leads you deeper into a dead end. Where exactly you stand—you may not know even at the end of your life.
The only advice I can give is this: observe whether the work brings you inner joy, even for a brief moment. If so, you will naturally keep doing it for a long time, and the question of “persist or give up” won’t even arise. Over time, the more hours you put in, the more experience you accumulate, the sharper your intuition becomes, the more efficient you get, and the more satisfaction you feel. A positive cycle emerges.
For me, thinking is enjoyable, and delving deeply into problems brings happiness—there is satisfaction every day I spend working on them. That alone is enough. Achievements are secondary, because they are not under your control. They always arrive late, and depend on others’ moods and subjective evaluations.
