About

I’m a final-year Ph.D. candidate in Machine Learning at the Georgia Institute of Technology, co-advised by Dr. Bo Dai and Dr. Chao Zhang. Most recently, I was a Student Researcher at ByteDance Seed in Summer 2026.

My goal is to build foundation models through the co-design of adaptive architectures and scalable learning systems, advancing efficient pretraining, test-time refinement, and agentic learning. In particular, I work on:

• Foundation model training — sparse pretraining with adaptive computation (EC-DIT), amortized language–image pretraining (AmorLIP), and architecture and scaling studies for Apple’s third-generation foundation models.
• Agentic learning & infrastructure — pluggable training infrastructure for end-to-end agentic RL (STACX), a learnable orchestrator–sub-agent harness for long-horizon ML engineering (Matryoshka Agent), closed-loop planning from feedback (AdaPlanner), and tool-using QA (ToolQA).
• Adaptive & self-improving systems — looped refinement for diffusion language models (FReDA), self-improving guidance for black-box LLMs (Matryoshka Pilot), test-time scaling of black-box LLMs through lightweight adapters (BBox-Adapter), and preference-aware context selection for personalized generation (HYDRA).

Before Georgia Tech, I received a dual engineer’s degree from CentraleSupélec (Paris) and a combined B.S./M.S. from Xi’an Jiaotong University through the Honors Youth Program.

In industry, I spent Summer 2026 at ByteDance Seed (San Jose) as a Student Researcher. Before that, I spent Summer 2024 and Summer 2025 – Spring 2026 as an AI/ML Research Intern on Apple’s Foundation Model Team, working with Dr. Nan Du, Tao Lei, Ruoming Pang, and Zhifeng Chen.

If you’d like to chat about research, collaborations, or just good coffee in Atlanta or the Bay Area, feel free to reach out at haotian.sun@gatech.edu.