About Me

I’m an incoming Ph.D. student in Computer Science at the University of Virginia, advised by Prof. Ferdinando Fioretto as part of the Responsible AI for Science and Engineering (RAISE) group. I’m honored to be fully supported by the UVA Computer Science Scholar Fellowship. I hold a B.Sc. in Mathematics from Beloit College.

My research focuses on constrained generative modeling to build multimodal AI systems that respect geometric, physical, and structural precedents. I’m interested in how constraints can be embedded directly into the generative process–through latent space design, decoding mechanisms, or optimization–so that outputs are more consistent, interpretable, and reliable. More broadly, I think of this as a step toward reasoning that reflects the structure of the real world.

I’m also interested in collaborative and open-sourced AI research, and I’m always happy to connect with others working on related problems.

Research Interests

I work on constrained generative modeling to build multimodal models that respect geometric, physical, and structural priors. My goal is to move beyond purely statistical reasoning toward outputs that are consistent, traceable, and reliable.

Geometry‑ and Physics‑Aware Generation – Enforcing geometric (e.g., symmetry, invariance) and physical constraints during training or inference of generative models (diffusion, flow matching, LLMs). This includes designing latent spaces and decoding mechanisms that guarantee constraint satisfaction.
Constraint‑Aware Optimization for Generative Models – Developing efficient training and post‑training methods (including RL‑based fine‑tuning) that embed constraints directly into the generative process. This approach reduces both inconsistency and computational cost.
High‑Stakes Decision Making – Applying constrained generative models to science and engineering, such as geospatial coordination, rural logistics, crisis response. These settings provide real‑world testbeds for reliability under hard constraints.

News

[Aug. 2026] Starting my Ph.D. in Computer Science at the University of Virginia as part of the RAISE group!

Publications

Le, Vu Anh and Dik, Mehmet, “A Mathematical Analysis of Neural Operator Behaviors,” Chapter 23 in Advances in Quantum Calculus and Functional Analysis, Taylor & Francis Group, July 2025.
Le, Vu Anh, Nguyen, Dinh Duc Nha, Nguyen, Phi Long, and Sood, Keshav, “RN-F: A Novel Approach for Mitigating Contaminated Data in Large Language Models,” in International Conference on Machine Learning Workshop on Data in Generative Models, June 2025.

A fun fact

Concentration of Measure Phenomenon

Why studying geometry is fundamental to the advances of multimodal AI? Let’s consider this fun fact. In high-dimensional spaces, distances between points become nearly similar. This reshapes how similarity and structure should be interpreted in learned representations. This phenomenon, named Concentration of Measure, is important to data geometry in AGI because it explains why naive notions of distance fail. That’s why it motivates the need for structured, constraint-aware representations that preserve meaningful variation.

Outside of work

Beyond academia, I’m an avid reader of history and philosophy. I’m exploring how diverse schools of thought can inform the design of aligned, augmented AI systems. I see common ground between Western analytic philosophy (with its naturalist worldview) and Taoism (with its concept of Ziran, or natural spontaneity). Though often viewed as opposites, both offer an observational, dissecting approach to learning from experience. I hope to enforce these principles and insights into long‑term scientific AI systems.