↓Skip to main content
Minkyoung Cho

Minkyoung Cho

Ph.D. Student in CSE at the University of Michigan

Google Scholar icon CV image/svg+xml CV CV

I am a fourth-year Ph.D. student, fortunate to be advised by Prof. Z. Morley Mao. My primary research interests lie in multi-modal, efficient AI, and robust AI. I’m particularly passionate about understanding what makes multi-modal models unique—especially through the lens of interpretability—by focusing on how to effectively and efficiently reconcile different modalities/agents. I also find great fulfillment in enhancing the efficiency & robustness of these models 😃!

News

GraphMap Accepted to IROS'26

GraphMap scales crowd-sourced global vectorized HD map construction through sparse visual graph fusion. See how GraphMap selectively fuses complementary observations across agents!

Internship at NVIDIA: Round 3!

Thrilled to return to NVIDIA for a third internship, this time joining the Autonomous Vehicle Research Group!!

Our Tutorial Accepted to ECCV'26

Excited to share that our tutorial has been accepted to ECCV'26! Stay tuned for more details.

DLA Accepted to ICML'26

DLA is a dynamic memory modeling framework for multi-state linear attention. See how it adaptively preserves important token information while keeping the state cache bounded!

TC-LoRA Accepted to NeurIPSW'25

TC-LoRA is a new paradigm that enables dynamic context-aware control by conditioning the model’s weights directly. See how we create and inject the appropriate conditional guidance at each step of the diffusion process.

SCORPION Accepted to IROS'25

We propose a robust, holistic 3D perception framework in a collaborative setting among multiple AVs. See how SCORPION mitigates temporal and spatial misalignments in feature fusion, even under lossy V2X communication!