About

About me

In English: Thu-Le Tran
In Vietnamese: Trần Thủ Lễ
Family name: Tran; Middle name: Thu; Given name: Le

My CV 2024.

Thu-Le Tran obtained his Ph.D. in Applied Mathematics from the University of Rennes, France, in December 2023. He served as a lecturer at Can Tho University since 2024. His research interests span from theoretical optimization, particularly in convex duality frameworks, to applications in:

1. Machine learning
2. Operations research
3. Signal/Image processing
4. Statistics

Since 2025, his research has been driven by several structural questions:

1. Static structure: What is the underlying structure of safe screening methods (feature vs. sample screening, static vs. dynamic screening, sequential screening, safe vs. unsafe rules, etc.)?
2. Dynamic structure: What is the underlying structure of optimization algorithms (gradient, proximal, greedy, and related methods)? Why are these algorithms so efficient across a wide range of optimization problems, particularly in neural networks?
3. Foundational structure: What is the underlying structure of convex analysis—its notions of convexity, operations, regularization, duality, optimality conditions, and algorithms—and how does this structure connect to related theories such as abstract and generalized convexity, matroid theory, Fourier and Fenchel transforms, idempotent analysis, and Lorentzian polynomials?
4. Meta-level structure: How should knowledge be structured and organized into an ontology? This includes representation learning, statistical learning, database models, training algorithms, explainable AI (XAI), Transformer models, and, more broadly, the development of sheaf-theoretic models of knowledge in which knowledge emerges from the interaction between an ontology (concepts and relations) and its realizations.

About my PhD:

My supervisors:

1. Cédric Herzet (PhD supervisor)
2. Clément Elvira (PhD co-supervisor)
3. Hong-Phuong Dang (PhD co-supervisor)
4. Valérie Monbet (Master supervisor)
5. Kien Trung Nguyen (Bachelor supervisor)

PhD Thesis: "Some Contributions on Safe Regions and Safe Screening in Convex Optimization". (PDF)

Abstract: Convex optimization is common in machine learning, statistics, signal, and image processing. Solving high-dimensional optimization problems remains challenging due to computational and storage constraints. In the last decade, the ``safe screening'' methods have become a powerful tool to reduce the dimension of these problems based on the knowledge of a ``safe region'' containing a dual optimal solution.

The first contribution of this thesis is a mathematical framework for creating new safe regions while demonstrating their superiority over the state-of-the-art. Our framework also provides an elegant way to unify existing safe regions. This contribution establishes a theoretical foundation for future advances in the study of safe regions.

The second contribution extends the safe screening methodology to problems in infinite dimensions. We show, in particular, that integrating this method into a state-of-the-art algorithm can significantly reduce its numerical complexity while preserving its convergence property. This contribution highlights the potential of safe screening to effectively address computational challenges in infinite-dimensional contexts.

About my contacts:

University Email: ttle@ctu.edu.vn

Personal Email: letranthu23041996@gmail.com

My webpages:

1. Google Scholar
2. List of lecturers at Can Tho University

Others:

1. Blogger, Wordpress
2. ResearchGate New, ResearchGate Old
3. Gitlab, Github, Github page
4. Scopus Preview, Orcid HAL
5. Facebook, Youtube

About this webpage

This webpage is my peronal blog aiming to update my latest publications, share what I learned and many other information.