About me: Hello world! I’m Quan Pham, an incoming PhD student at the National University of Singapore. I received my bachelor’s degree from HUST. I was mentored by Prof. Trung Le and Prof. Dinh Phung (previously at VinAI Research), and I currently work closely with Dr. Tung Pham and Dr. Hung Bui at Qualcomm AI Research.

I love experimenting with language models to better understand the nature of their emerging abilities. I’m especially curious about the theory behind LLM abilities: what explains the behaviors we see, and how we can make them more reliable.

When I’m not doing research, you’ll probably find me building LEGO, watching films or off traveling somewhere new. I like collecting moments and photos along the way, and you can browse a few of them in my travel gallery.

Updates

• [Feb 2026] I was offered PhD admission and research scholarship from National University of Singapore.

• [Dec 2025] I am actively looking for a PhD position. Good luck to me!

• [Jun 2025] I will attend ICML 2025. Let’s connect!

• [May 2025] I got 2 papers accepted at ICML 2025.

• [Apr 2025] I joined Qualcomm AI Research as an AI Resident.

• [Dec 2024] I achieved an IELTS Academic band score of 7.0!

• [Aug 2024] I joined VinAI Research as an AI Resident.

• [Apr 2024] My first paper was accepted at ICASSP 2024.

Research Interests

My research lies at the intersection of statistical machine learning and large language models (LLMs). I am interested in developing principled foundations for understanding how overparameterized models learn, generalize, and reason, as well as in designing probabilistic inference frameworks that fully leverage the intrinsic capabilities of LLMs.

A central theme of my work is to bridge statistical theory and modern foundation models. In particular, I focus on:

• Developing probabilistic inference methods for foundation models, with an emphasis on improving reasoning capabilities in LLMs.
• Advancing scientific understanding of LLMs by investigating the fundamental statistical properties that govern how models learn and reason.