The research group at University of Texas at Austin published a paper titled “Convergence Rates for Latent Mixing Measures in Infinite Homoscedastic Location-Scale Mixture Models” last week on the arXiv platform.
The study focuses on the problem of information limits in Bayesian nonparametrics. Dung, currently a second-year PhD student at the university, is the lead co-author under the supervision of Ho Pham Minh Nhat, an associate professor there.
Le Quang Dung, a PhD student at the University of Texas. Photo courtesy of Dung |
The paper is the first in a planned series of 10 studies aimed at fully resolving a major open problem in Bayesian nonparametrics dating back to the 1970s.
“This research uses tools from pure mathematics, but it is crucial for understanding models used in statistics. Solving this problem could lead to many other interesting results,” Dung said.
According to the authors, the problem involves extracting hidden parameters in machine learning data, including central tendency and dispersion. Previous studies typically assumed dispersion was already known in order to solve the problem. However, when both parameters are unknown in real-world settings, traditional mathematical tools fail completely, leaving the problem unresolved for many years.
The “super-smooth” nature of multivariate Gaussian distributions and similar kernels acts as an information sink in data science, signal processing, and machine learning. Attempts to reverse-engineer these parameters cause statistical noise to grow exponentially, making accurate machine learning extremely difficult.
To overcome the challenge, Dung and his colleagues combined advanced tools from multiple branches of mathematics. His approach is notable for its flexible use of techniques from functional analysis, Fourier analysis, and distribution theory.
While these techniques are classical and familiar to mathematicians, recognizing their connection to the problem offered a new perspective that helped resolve its core challenges. Once the direction became clear, the team completed the work in five months.
“Dung has an unwavering commitment to mathematical rigor, never compromising with approximate arguments or loose conjectures. Every obstacle is examined thoroughly and resolved with meticulous proof line by line,” Nhat said.
arXiv is a globally recognized online repository for preprints that allows scientists to share research early and establish intellectual ownership before formal journal publication. Maintained by Cornell University, the platform says submissions are screened by leading scientists with top academic qualifications in their respective fields.
Dung is a former mathematics student at Lam Son High School for the Gifted in Thanh Hoa Province. During high school, he won two first prizes in Vietnam’s national mathematics competition and earned a gold medal at the 2017 IMO.
He graduated with distinction from the mathematics honors program at Vietnam National University, Hanoi’s University of Science in 2021 before receiving a scholarship to pursue a master’s degree at École Polytechnique in France and later continuing his studies in the United States.
Comments are closed.