Male student researches to improve AI's 'biased learning'

Passion for AI started in 11th grade

Hai said he learned about and became passionate about AI research from a chance conversation with his chemistry teacher when he was in 11th grade. The teacher introduced him to AI and opened a new door for him. With a passion for discovery, Hai decided to enroll in the data science and AI major at Hanoi University of Science and Technology, a major that requires high entrance requirements and is challenging.

Here, Hai said he experienced a solid mathematical foundation training program in the first two years, which made it easier to approach and pursue AI. In addition, Hai also participated in AI research labs, where he learned and developed skills from students and lecturers.

Nguyen Nam Hai. PHOTO: NVCC

In 2024, Hai received a scholarship for students with excellent grades in the semester, and successfully applied to the residency program of VinAI (AI Research Talent Training Program, under Vingroup) which was officially acquired by Qualcomm Company since April 1.

Earlier this year, Hai’s first paper, “Distributed-Centric Fairness Computing via the Sliced ​​Wasserstein Method,” was accepted at the ICLR 2025 Conference (the premier conference in the field of AI). Hai’s paper was also featured in the spotlight (the top 3.2% of papers) at the conference.

Hai said he joined VinAI when he was a third-year student, and is the youngest AI research intern in the company to date. When he first started working, Hai felt pressured to balance between studying and working, but after a while, he found a way to integrate and manage his time effectively.

"Many days I was working at the company, had to take the bus to school to be on time for class and vice versa. As time passed, I gradually got used to that harsh intensity and was able to balance my time in class as well as my research at the company," Hai recounted.

Improving AI's "biased learning"

Hai started his research with the goal of building a fair and efficient AI system. "The main content of the paper comes from a practical problem. When AI learns from unbalanced data, the results are often biased. For example, I want to create an AI model that generates images of digits from 0 to 9. If the input data has too many 0s and 1s, the model will learn biasedly and generate mainly these two numbers, ignoring the other numbers," Hai shared.

Hai is considered to have a receptive spirit and sharp analytical thinking. PHOTO: NVCC

Hai said the team proposed a new method to find a “balanced center point” between data sources. This approach ensures that all components are treated equally. As a result, the generative model will produce digits with equal frequency, without favoring any one number.

One of the biggest challenges Hai faced during his research was the ability to keep up with the rapid development of the industry. "I taught myself through books, read scientific reports and participated in research projects at VinAI to access new technologies," Hai shared.

Hai is preparing a scientific paper to submit to the ICML 2025 Conference, researching a method to compare two datasets at superior speed, contributing to optimizing AI training.

"The AI ​​industry in Vietnam is growing strongly and opening up many career opportunities. However, competing with international corporations is also a big challenge. I need to constantly learn and update trends to keep up with the development of technology," Hai said.

With the successes achieved, Hai's next goal is to continue to deepen research in the field of AI and complete his application for a PhD. Hai also wants to learn more about AI and solve practical problems in the fields of education and medicine.

Hai has been a member of the research team led by Dr. Ngo Van Linh, a lecturer at the School of Information and Communication Technology, Hanoi University of Science and Technology, for 3 years. Dr. Linh commented: "Hai is enthusiastic, creative and disciplined. During the research process, Hai always demonstrates sharp analytical thinking and admirable self-learning ability. In terms of personality, Hai is modest, honest and responsible. Despite the great pressure of work, he always completes on time and is willing to support other members of the team."

Dr. Linh added: "What I appreciate most about Hai is his open-mindedness, his willingness to accept feedback, and his ability to see each challenge as an opportunity to grow. I believe that with a solid foundation and professional working attitude, Hai will continue to shine in his AI research journey. I fully support and confidently recommend Hai for scholarship programs, research projects, or positions requiring high expertise."

Source: https://thanhnien.vn/nam-sinh-nghien-cuu-cai-thien-viec-hoc-lech-cua-ai-185250406150356788.htm