Floria Jin

I am a B.S. student in Statistics (Minor in Mathematics) at the University of Illinois Urbana-Champaign, graduating in 2026. I think of myself as a junior AI researcher working at the intersection of statistics, machine learning, and AI safety.

My central interest is AI safety and alignment—making large language models and agentic systems more robust, honest, and controllable. I have reproduced and studied a range of alignment methods, including SecAlign (prompt-injection defense), Circuit Breakers (representation engineering), Safe-RLHF, and Latent Adversarial Training, and I read closely in agent safety, multi-turn risk, and code-agent security. Alongside this, I work on causal inference and causal ML, Bayesian modeling and optimization, and graph machine learning.

What I rely on most is cross-domain transfer—an instinct for connecting ideas that look unrelated. I tend to take a research framework apart functionally, ask what each component does and where it transfers, run a small experiment that alters the causal direction, and then combine complementary streams into a single working loop. That habit has carried across counterfactual climate-risk modeling at City University of Hong Kong, a memory system for a mental-health support agent at Shanghai Jiao Tong University, and AI/software engineering internships at NetEase Youdao, PhinD, and in finance (graph-based fraud detection).

Before statistics and computer science, I studied the social sciences—and I think of that training as a gift I carry into the technical work. You can read why in Another River.