Jinqi Luo
I am Jinqi Luo, a CS PhD candidate in the 
University of Pennsylvania. I am advised by Prof. René Vidal, and my thesis comittee chair is Prof. Chris Callison-Burch. I obtained MS in Robotics (MSR) in the Robotics Institute of 
Carnegie Mellon University. I was affiliated with CMU Computer Vision Group, advised by Prof. Fernando De la Torre and Dr. Dong Huang. I obtained the BEng in Computer Science (Highest Distinction) at 
Nanyang Technological University (NTU Singapore), advised by Prof. Jun Zhao. I have experiences at 
Amazon (Rufus MLLM), 
Alibaba (Alipay), UC Berkeley (visiting student), and Duke University (RA).
Research Interests
My work centers on a core research problem: how do we develop intelligent systems that are controllable, compositional, and aligned? My past papers discover solutions of representation steering, low-rank adaptation, sparse optimization, and robust learning for generative foundation models such as LLMs (NeurIPS 2024, ICML 2026), MLLMs (CVPR 2026), Diffusion Models (CVPR 2025), World Models (ICCV 2025, NeurIPS 2024), and GANs (CVPR 2023, AAAI 2021). Some of my work positions unified alignment paradigms for deep learning (TMLR 2025, IJCAI 2021). Please feel free to reach out to discuss potential collaborations and opportunities.
Recent Works (full list click here)
Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs
Concept Lancet: Image Editing with Compositional Representation Transplant
PaCE: Parsimonious Concept Engineering for Large Language Models
LogiCity: Advancing Neural-Symbolic AI with Abstract Urban Simulation
Recent Advances in Adversarial Training for Adversarial Robustness
ActivityGAN: Generative Adversarial Networks for Data Augmentation in Sensor-Based Human Activity Recognition
Generating Adversarial yet Inconspicuous Patches with a Single Image
Services and Activities
- Conference Reviewer: CVPR, NeurIPS, ICLR, ICML, ICCV, EMNLP, ACL ARR, AISTATS, AAAI, ECCV.
- Journal Reviewer: TMLR, Neurocomputing, TNNLS.
- Organizer: ICLR FM-Wild Workshop.
- Teaching Assistant: [Penn 23Fall] Deep Generative Models (Founding Members). [Penn 24Fall] Deep Generative Models.