Nam Cao

M.S. Student in AI

KAIST AI · OSI Lab

I am currently seeking Ph.D. or Research Engineer positions.

My research centers on efficient LLMs [C1, P1] and high-performance inference and serving systems. Recently, I have been interested in speculative decoding with diffusion language models as drafters [C1], a path toward faster and more cost-efficient inference without sacrificing quality under constrained compute budgets. In parallel, I have been exploring AI safety for LLMs [C2], as well as AI for Science, currently through foundation models for weather forecasting.

Education

M.S. in Graduate School of AI, KAIST. Advised by Prof. Se-Young Yun Feb 2025 – Feb 2027 (expected)
B.S. in Computer Science, KAIST. Aug 2020 – Aug 2024

News

Sep 2026 🎉 BASTION has been accepted to NeurIPS 2026. This is my very first co-first-author paper at a top-tier conference!
Sep 2026 🎉 SAGA has been accepted to NeurIPS 2026.
May 2026 🎉 BASTION selected for an oral at the ICML 2026 AdaptFM workshop.
2025 🏆 Awarded the KAIST Global Presidential Scholarship (KGPS).
Feb 2025 🏫 Started my M.S. at KAIST AI, joining OSI Lab under Prof. Se-Young Yun.
Aug 2024 🎓 Graduated from KAIST with a B.S. in Computer Science.

Publications

^* equal contribution · C conference W workshop P preprint

2026

[C1]BASTION: Budget-Aware Speculative Decoding with Tree-structured Block Diffusion Drafting

Soowon Oh^*, Nam Cao^*, Yujin Kim, Hojung Jung, Huzama Ahmad, Sangmin Bae, Se-Young Yun

NeurIPS 2026 Poster

Earlier version: AdaptFM Workshop, ICML 2026 Oral

arXiv Code
[C2]Stage-wise Attention-Guided Region Sequencing for Adversarial Attacks on Large Vision-Language Models (SAGA)

Jaehyun Kwak, Nam Cao, Boryeong Cho, Segyu Lee, Sumyeong Ahn, Se-Young Yun

NeurIPS 2026 Poster

arXiv
[P1]CascadeFormer: Depth-Tapered Transformers Motivated by Gradient Fan-in Asymmetry

Huzama Ahmad, Nam Cao, Se-Young Yun

Preprint, 2026

arXiv

Research Projects

Weather Foundation Model Aug 2026 – Present

National Institute of Meteorological Sciences (NIMS), Korea × OSI Lab, KAIST

Building a generative weather foundation model trained on ERA5 for forecasting from incomplete input.
On-Device Speculative Decoding Feb 2026 – Aug 2026

Samsung Research (SoC Architecture team) × OSI Lab, KAIST — Co-Project Leader

Developed BASTION, a hardware-aware speculative-decoding framework with block-diffusion drafters, for faster on-device LLM inference.
Deep Learning for Air-Quality Forecasting Feb 2025 – Mar 2026

National Institute of Environmental Research (NIER), Korea × OSI Lab, KAIST

Built DustViT, a vision-transformer model for short-term particulate-matter (PM) forecasting.

Awards & Honors

KAIST Global Presidential Scholarship (KGPS) — Four-semester graduate scholarship with a monthly living stipend. (2025–2027)
KAIST–Qualcomm Innovation Award — Winning team at the KAIST–Qualcomm AI Hackathon. (2023)
KAIST International Undergraduate Scholarship — Full-ride scholarship. (2020)
Vietnam Mathematical Olympiad (VMO) — Bronze Medal. (2018, 2019)
LLM Agents MOOC, Ninja Tier — UC Berkeley RDI. (Fall 2024)