Reflection: My First Research Experience

In 2019, I started my first research experience with Dr. Yupeng Li. It is not perfect, but I am always grateful to Dr. Yupeng Li, who guided me to the research career.

First Meet in AiTrans Competition

I met Dr. Li the first time I participated in the AiTrans Competition at the end of 2018. It is a competition about AI-enhanced data transmission. Because of the network fluctuations, users will have a terrible experience if they watch videos with low network speed. Researchers adopt AI methods to automatically adjust the video’s resolution based on the network’s history information to provide a fluent user experience to solve the problem.

As I have mentioned in Reflection: Being a Software Engineer, I have learned some basics of machine learning and deep learning in the fall of 2018, so I joined this competition with Mr. Senyue Hao and Mr. Ziheng Mo. Although it provided an RL-based solution as the reference, we had no experience of RL before. Consequently, I developed a search-based algorithm for this problem and implemented it in Python. After weeks of debugging and testing, we entered the final competition at Tsinghua University and won the best presenter prize in the final round.

During this competition, Dr. Li was our supervisor. We had met several times before the final, and he suggested us to keep improving our algorithm. I found he was a nice guy, so we met again after the competition to discuss his research topics.

Reference link: https://mp.weixin.qq.com/s/hJuYg9BkW76f6X92u9uf7Q

Left: Dr. Yupeng Li; Left-2nd: Mr. Senyue Hao; Right-2nd: Me; Right: Mr. Ziheng Mo

Another Outcome: One patent accepted: https://patents.google.com/patent/CN110087110A/zh

Ride-Sharing Project

Dr. Li was working on some ride-sharing projects cooperating with Didi Chuxing at that time. Ridesharing was still a novel and rising topic in 2019. In the spring festival, he assigned several papers to explore, including the scheduling and routing algorithms of ridesharing. After the spring festival break, he assigned me to the topic of activity-based ridesharing.

Activity-based ridesharing is a novel business model which is not implemented in reality yet in 2021. Instead of indicating the exact destination, passengers will tell their activities like go shopping or see movies. Such modification provides more potential for routing optimization, like grouping passengers to the same goal to save money and energy.

My task is to develop a new type of activity-based ridesharing and propose an algorithm for it. I changed the objective from the energy or money saved to the activity’s property and formulated an Integer Programming problem. Although it is a well-defined formulation, the major challenge is the difficulty and the contribution of this problem. Dr. Li required me a perfect solution with a theoretical upper bound and avoided the heuristic ones like ML, so I suffered months studying randomized algorithms.

Personally speaking, it is very tough for me to learn the randomized algorithms by myself. It is a well-studied problem, but there is no good textbook because it is not a hot area like ML. I read lots of math papers with complex formulas everywhere, and I gradually lost my confidence.

Finally, I proposed a very rough solution with a useless upper bound (in my opinion), and I had decided to switch to CS systems research, so I gave up this project after a long meeting with Dr. Li in the spring of 2020. Thanks for his understanding; we are still friends and still keep in touch. In the next month, I completed my system’s research and submitted my paper.

One outcome of this research is two patterns submitted by Didi Chuxing. In my opinion, the most important lesson I learn is the research method. By experiencing the process of literature review, brainstorming, problem-solving, experiments, and paper writing, I understand what research is. Moreover, by learning randomized algorithms, I know that I still have a lot to learn in math, and I dislike pure-math optimization problems.

If You Want to Do Research

The most important thing is to get started as soon as possible. I have witnessed many students wait and wait and finally waste their undergraduate time.

Try to reach out to professors and talk with them. Usually, the first research may not succeed, but you will always learn something and have more options in the future.

Read More

To read more, you may go back https://blog.xieguochao.com/2021/03/26/decided-epfl-mscs/ .

1 thought on “Reflection: My First Research Experience

Leave a Reply

Your email address will not be published.