Peilin Cai’s Personal Website

About

I am a master student researcher focused on computer vision (CV), large language models (LLMs), and multimodal generation. At USC’s Graphics & Vision Lab (advisor: Prof. Yue Wang), my work centers on 3D reconstruction under sparse observations, controllable generative rendering, and embodied navigation. More broadly, I explore the intersection of generative models, world modeling, and embodied intelligence: how to build interactive, explorable, high-fidelity worlds from very small sets of real images; how to couple geometric priors with diffusion/autoregressive models to produce videos with temporal consistency and realism; and how to make these capabilities run reliably on edge and in-vehicle platforms.

Before this, I carried out two research projects of great personal significance at Prof. Yue Zhao’s FORTIS Lab: SecDOOD (ICCV 2025 Poster) and PERSONABENCH (NeurIPS 2025 MTI-LLM Spotlight). The former proposed a secure on-device OOD detection framework that requires no gradient backpropagation, offering insights for deploying personalized large models on edge devices; the latter introduced the first benchmark for evaluating the personalization capabilities of LLMs in multi-turn conversational settings. I am deeply grateful to Prof. Yue Zhao and the senior PhD students in the lab for their support.

I have extensive experience in multimodal out-of-distribution (OOD) detection and in the training, deployment, and inference of LLMs, and I am also honing my research skills in computer vision and robotics at the GVL Lab. My primary research interests include probing and analyzing the limitations of LLMs, embodied intelligence and visual generative models. If you are interested in collaborating, please feel free to reach out. My preferred email is peilinca@usc.edu

Current State:

• In my third semester of the M.S. in Computer Science program at the University of Southern California.
• Currently working on two projects that will be submitted for publication.
• Seeking PhD opportunities.

Publications

CV