Park Jae-sung, a 4th-year student in the Department of Civil and Environmental Engineering, and Kim Do-hyun and Han Ji-min, both 4th-year students in the School of Data Science at Hanyang University, received the Grand Prize and Encouragement Award, respectively, at the Hecto 토토사이트 무소유 Challenge held this past July. The competition tasked participants with developing 토토사이트 무소유 models capable of classifying used car models based on real-world images. We spoke with the students to learn more about their journey through data analysis, model building, and real-world problem solving.

Park focused not on using t토토사이트 무소유 most powerful model, but on selecting one optimized for t토토사이트 무소유 specific requirements of t토토사이트 무소유 task. 토토사이트 무소유 began by analyzing t토토사이트 무소유 characteristics of t토토사이트 무소유 dataset, identifying noise and irrelevant images that could cause misclassification. Through hands-on experimentation with various open-source models and visualization results, 토토사이트 무소유 ultimately selected t토토사이트 무소유 most suitable architecture.

During the competition, Park encountered an issue where identical code produced different results depending on the GPU architecture. “It wasn’t a simple coding error,” he explained. “It stemmed from execution differences in the hardware environment.” By analyzing the GPU’s open-source components himself, he was able to pinpoint the exact source of the problem. “This process allowed me to go beyond just using technology—I was able to truly understand what’s happening inside,” he reflected.

Park, who first became interested in programming during middle school, is now expanding his career path by studying entrepreneurship major. “There’s a crucial need for people who can bridge the gap between users and technology,” he said. “This competition gave me the opportunity to further explore that role.”

Park aims to become a “practical AI planner,” one who integrates text-based AI models into industrial fields and helps bridge the gap between research and real-world application. “In this fast-changing technological era, it’s important to find your own niche,” he shared. “Rather than focusing solely on the technology itself, I want to define and interpret problems at the intersection of people and technology.”

Teammates Kim Do-yeon and Han Ji-min focused on data analysis for their challenge. “It was hard to get the model to train reliably because used car images vary greatly in lighting and angles,” Han explained. “From the beginning, we tried to reflect conditions similar to real industry settings.”

“There were many visually similar vehicles, which made it necessary to extract features with greater precision,” Kim added. To address this, they applied data augmentation techniques such as cropping, flipping, and color adjustments to increase data diversity, along with fine-grained classification strategies to distinguish between similar models. They also utilized contrastive learning techniques to improve model representation.

They admitted to feeling anxious when their model underperformed mid-competition. “There were moments when I doubted my entire design after a single drop in score,” Han recalled. “But having teammates to ask questions and offer suggestions from different perspectives gave me strength.” Kim added, “Dividing roles based on our strengths really helped.”

Their undergraduate experience also played a key role. Han noted, “Working as an undergraduate research assistant for a year helped me a lot—reading papers and presenting in seminars trained my analytical thinking, which I applied here.” Kim emphasized that her prior experience with code reviews and experimental design directly influenced their success during the competition.

The students are interested in more than just developing models—they aspire to build systems where AI can think as flexibly as humans. “I plan to pursue research in the field of multimodal AI,” said Han. “It’s about building AI that naturally blends into everyday life by understanding images, text, and audio together.” Kim, for her part, is focused on designing systems that integrate various input types, particularly through connections with large language models (LLMs).

Han left a message for students preparing for a career in AI: “Rather than just chasing the right answers, the ability to ask your own questions is what really matters. If you study AI through your own interests and approach, you’ll eventually find your own answers.” Kim added, “If you find value in the challenge itself and keep stacking up experiences, you’ll surely carve out your own space.”