Welcome

Thank you for visiting my website. This site is primarily intended to present my curriculum vitae, but it also features content such as notes from my studies, brief introductions to my research projects, and a playground with lightweight web applications related to my academic work. You can navigate the site using the text buttons at the top of the page:

  • If you are looking for my curriculum vitae, click the "JUN SU PARK” button at the top left
  • If you are interested in my studies and research, use the text buttons or the dropdown menu at the top right

I hope you find the content here useful and insightful. If you would like to get in touch, feel free to email me at junsu@yonsei.ac.kr

Thoughts

This may come a little unexpectedly, but I would like to share some personal thoughts that have emerged from my recent academic work. I hope this helps me connect with others who think similarly, and possibly opens doors to future research collaborations.

  • I believe that today’s deep learning models perform remarkably well. However, at their core, they are still advanced tools that carry out nonlinear, high-dimensional mapping. They do not yet achieve real human-like thinking or understanding. Because of this, I see clear limitations in their ability to solve problems that are complex, open-ended, or different from the data they were trained on.

  • To overcome these limitations, I believe it is not enough to simply improve deep learning models. We need to combine deep learning with other kinds of artificial intelligence and mathematics. For example, we may find more powerful solutions by combining different problem-solving approaches, such as:

    • Hard computing and soft computing
    • Rule-based and data-driven decision-making
    • Deterministic mathematical models and deep learning models

  • Some people believe that, in the end, all problems can be treated as mapping problems, no matter how complex or open-ended they seem. If this idea is true, then deep learning might one day evolve into a form of general intelligence. If that happens, the combined approaches I mentioned earlier may no longer be necessary. But at this point, we do not know if this idea is right or wrong. That is why I believe we should continue using a wide range of tools including deep learning, other artificial intelligence methods, and mathematics to solve real-world engineering problems. At the same time, we should keep exploring whether this idea about mapping truly holds.

  • If, on the other hand, this idea is not true, and not all problems can be treated as mapping problems (which currently seems more likely), then the need for hybrid approaches becomes even more important. In that case, we must also continue searching for ways to build general artificial intelligence that can deal with complex and open-ended engineering problems. To succeed, we will need new and original ideas that go beyond the current paradigm of deep learning. I believe these ideas could come from anyone, regardless of their background. That is why I believe it is meaningful for all of us to reflect deeply on these questions.

My long-term research goal is to explore and find answers to these fundamental questions. Along the way, I also hope to develop practical and advanced solutions in my primary field, architectural engineering. Currently, I am focusing on two main topics:

  • Developing methods for single-agent design optimization for architectural engineering
  • Developing methods for vision-based remote inspection and monitoring for architectural engineering