Amrit Singh Bedi

I am an Assistant Professor in the Department of Computer Science at the University of Central Florida, where I lead the SafeRR AI Lab. The modern generative AI development cycle has three stages: pre-training, post-training, and deployment. Our lab focuses on the latter two, targeting cutting-edge problems across LLMs, VLMs, vision-language-action models, diffusion LLMs, and agentic AI.

At the post-training stage, we develop novel fine-tuning and reinforcement learning methods that align AI systems with human values and safety constraints. At the deployment stage, we design inference-time algorithms that make models safer, more robust, and more reliable — without retraining. Our key research themes include:

• — AI Alignment — RLHF, preference learning, and reward modeling
• — Safety & Trustworthiness — inference-time safeguards, jailbreak defenses, and robustness in generative AI
• — Reinforcement Learning — multi-agent RL, hierarchical RL, and embodied intelligence
• — Optimization & Theory — bilevel, non-convex, and federated methods for modern ML

Prior to UCF, I held positions at the University of Maryland, College Park, and the U.S. Army Research Laboratory. I received my Ph.D. in Electrical Engineering from IIT Kanpur.