Le Danh Vinh, valedictorian of the elite computer science talent program at Hanoi University of Science and Technology, had a lead-author paper accepted to ICML 2026 in Seoul and is a co-author on a paper accepted to EMNLP 2026 in Budapest.
Vinh graduated with distinction in late September with a GPA of 3.92 out of 4. He works at VinMotion, Vingroup’s humanoid robotics company, and is a member of DreamAI, a research group founded by Trung Le, a machine learning researcher at Monash University in Australia. “It’s something I never imagined in my early days at the university,” he said.
A former student of the specialized math class at Bac Ninh High School for the Gifted, Vinh was admitted to the university without an entrance exam in 2022 after winning second prize in the national excellent student contest in mathematics. He then passed an interview to enter the talent program.
From his first year at university he went beyond set textbooks and lecturers’ reading lists to study specialist material on machine learning and deep learning, subjects not formally taught until the third year. He scored a perfect 4.0 GPA in his first two semesters and won first prize in analysis at the national Mathematical Olympiad for university students, by which time he had settled on a path in artificial intelligence.
His search for internships and lab positions, which began in the second semester of his second year, initially met with little success. His research career took shape in his third year when he joined ZenAI, a generative AI startup, where he led several projects. Lecturer and adviser Ngo Van Linh gave him the chance to do research alongside him.
Vinh’s research focuses on efficient AI — methods that cut the cost of running models and speed up inference — and on vision-language models (VLMs), which combine information from images and text. “I’m also interested in developing more efficient models and using VLMs for vision-language tasks, as well as applications in humanoid robots,” he said.
ICML acceptance and competitive context
The final push on his ICML 2026 paper coincided with end-of-semester exams and a full workload at VinMotion, and he faced a shortage of computing hardware for experiments. “Some days I slept four hours or less because I wanted the paper to be accepted so badly, but I couldn’t neglect my work at the company or my exams at school,” Vinh said. ICML accepted the paper, on which he is first author. The conference, rated A* in the CORE ranking, was held in Seoul in July and turned away more than 73% of a record of some 24,000 submissions, according to its organizers.
Vinh is also a co-author of a paper accepted to EMNLP 2026, a leading natural language processing conference scheduled for Budapest in late October.
Linh, who has worked with Vinh for nearly two years, said he was struck by Vinh’s independent thinking and persistence. When Linh set him a new direction — applying knowledge distillation to resource-hungry multimodal embedding models — early results were inconclusive, but Vinh kept working through the problems. When ICML reviewers raised tough questions, Vinh built his own arguments and verified them convincingly without waiting to be guided.
At VinMotion he is looking for new challenges and plans to gain more work and research experience before pursuing graduate studies.
