Nghia H. Nguyen

I am a PhD student in the Computer and Information Science Department at the University of Pennsylvania, advised by René Vidal.

My research focuses on the emergence and utilization of compositional structure in foundation models for trustworthy machine learning. Rather than treating latent spaces as unstructured black boxes, I study how models encode compositional primitives (from hierarchical semantic taxonomies to algorithmic reasoning) and pursue two core questions:

• Principled concept recovery & verification: How can we exploit hierarchical and compositional priors in representation space to extract reliable, faithful explanations and defend models against semantic vulnerabilities? (e.g., HCEP).
• Learning dynamics & structural emergence: Under what conditions do training objectives induce compositional geometry? I work on proving how multi-modal and self-supervised training dynamics translate compositional structure in data into geometric properties in representation space (e.g., HAS).

Previously, I did my B.S. in Computer Science at the University of Science, Vietnam National University – Ho Chi Minh City, advised by Minh-Triet Tran. I was also in the AI Residency program at FPT Software AI Center, working with Thieu N. Vo and Tan M. Nguyen. In Summer 2020, I was a Research Intern at the Distributed Information Systems Laboratory (LSIR), EPFL, working on Graph Neural Networks for SQL query optimization.

If you have any feedback or suggestions on my research, talks, or collaborations, feel free to leave anonymous feedback!