gianghle Goto Github PK

I'm an AI researcher who is interested in computer vision, natural language processing, and generative models.

You can find more about me on my Stack Overflow, DACON account, and LinkedIn.

Some things I wanna share:

• I earned a Master's degree from the Seoul National University of Science and Technology (SNUT). Yes, I know, it's not SNU, it's SNU-T.

• During my master's degree, my field of study was generative models with a main research focus on GANs. After StyleGAN3 was published, I wished I could work with Tero Karras to see how he came up with those ideas.

• Besides, I believe the diffusion architecture shows more promising results than the GANs approach, thanks to the fine-grained control in training instead of relying on true/false feedback from the discriminator. (no more mode collapse)

• I first discovered the word "machine learning" in 2019. When I was at the beginning of my third year of bachelor's, one of my best friends suggested we take this course (it used to be on Coursera). We spent 8 to 12 hours a day, continuously for 14 days, to finish this course, covering all lectures, videos, notes, and assignments (in Octave).

• Then I had my first intern job, which wasn't ideal, but still a valuable experience. My friend and I then completed a significant project on monitoring traffic for our graduation. Project details can be checked here.

• Currently, I'm diving into NLP to gain a better understanding and to participate in some competitions on Kaggle. Additionally, I believe that multi-modal learning is the future of machine learning, although it might not be prevalent in this era. I think models that focus on specific tasks are currently more valuable in the market.

Note: Since I noticed a lack of standard packages for GANs evaluation metrics such as IS, FID, Precision, and Recall, I implemented one myself. You can check the details at this pip link. Feel free to use it or raise any issues if you find any mistakes in my implementation.

I love joining competitions and reading solutions to learn how to handle machine learning models with real-world data. Since I don't have enough resources to deal with big data, platforms like DACON are more suitable for me at this time. Here are some of my achievements:

• Camera Image Quality Improvement AI Competition, hosted by LG AI Research and organized by DACON, I participated solo. Out of 228 teams, I reached 10th place, which placed me in the top 4%. The task involved developing an AI model to improve camera image quality degraded by light blur.

• Monthly Daycon Computer Vision Outlier Detection Algorithm Competition, organized by DACON, my labmate and I participated. We reached 13th place out of 480 teams, placing us in the top 3%. The task involved the development of a computer vision algorithm to classify the type and state of objects.

• The 2nd KRX Stock Investment Algorithm Competition, hosted by Korea Exchange (KRX), sponsored by Koscom, and organized by DACON, I participated solo. I reached 41st place out of 400 teams, placing me in the top 10%. The task involved using capital market data and public data to create a stock investment algorithm capable of expecting high stability and returns.

• Handling questions and answers about wallpaper defects: Hansol Deco Season 2 AI Contest, hosted by Hansol Deco (한솔데코) and organized by DACON, I participated solo. I reached 18th place out of 557 teams, placing me in the top 3%. The task involved developing an AI model with in-depth question-and-answer processing capabilities related to wallpaper.