Hyunwoong Bae
Ph.D. Candidate, M.IN.D Lab, Seoul National University
Research Interest: Model perception and understanding
Ongoing Research: Byte-level Transformer, Robotics safety with VLLM policy
Ongoing Research: Byte-level Transformer, Robotics safety with VLLM policy
Education
2022.09 – Present
M.S.–Ph.D. Integrated Program in Artificial Intelligence, Seoul National University
2016.03 – 2022.08
B.S. in Semiconductor Systems Engineering, Sungkyunkwan University · Summa Cum Laude
Preprint
P-Harness: A Harness for Answer Generation over Personal Memory
Preprint
Benchmarking Obstacle-Conditioned Safety in Household Robot Navigation
IEEE IROS 2026 Workshop SeMaNa
SafeHRI-Nav: A Context-Aware Low-Level Safe Navigation Benchmark for Human-Robot Interaction
EMNLP Main 2026
Axis-LoRA: Probing and Adapting Embedding Models for Personal-Memory Retrieval
ICCV 2025 Workshop SafeMM-AI · non-archival
Revisiting VLLM Safety Evaluation: Disentangling Benign Grounding from True Safety Failures in VLLMs
Microscopy and Microanalysis 2025
Efficient and Robust SEM Image Denoising for Wafer Defect Inspection
* Equal contribution
2026.09 – 2026.12
Teaching Assistant (Electrical and Electronic Technologies in Industrial Applications)
Dept. of Electrical and Computer Engineering, Seoul National University
2026.07 – 2026.10
Teaching Assistant
Samsung Electronics AI Expert Program
Mentored an on-the-job project predicting post-bond wafer quality and defects at the pre-bond stage in hybrid bonding, by combining metrology images with leveling sensor data.
2024.05 – Present
Lab Server Management
M.IN.D Lab, Seoul National University
2023.09 – 2023.12
Teaching Assistant (Basics of Deep Learning)
Dept. of Electrical and Computer Engineering, Seoul National University
2022.09 – 2023.08
Industry-Academic Cooperation Lead
Samsung Electronics × Seoul National University
Denoising strategy to replace conventional methods while maintaining semiconductor-related inspection metrics.
2022.01 – 2022.02
Undergraduate Research Assistant
Architecture and Code Optimization Lab, Seoul National University
Optimized the Deep Learning pipeline of NVIDIA DALI by balancing IO & CPU
2021.07 – 2021.08
Intern
Examined & patched the kernel of AndroTrace, which traces and analyzes Android IO
2020.10 – 2021.11
Undergraduate Research Assistant
Scalable Architecture Lab, Sungkyunkwan University
Neural Network–based Packet Recovery System for Real-time Collaborative Intelligence over Unstable UDP Networks
2024.09 – 2025.02
Scholarship of Jung-hun Foundation
2022.09 – 2024.08
Scholarship of AI Fellowship
Seoul National University
2022.08
Summa Cum Laude (성균 최우등 졸업상)
Sungkyunkwan University
2019, 2021
Dean's List, College of Information and Communication Engineering
Sungkyunkwan University
2019.12
2nd Prize in Co-Deep Learning Project, Center for Innovative Higher Education
Sungkyunkwan University
Design and Implementation of a Drone Propeller-Driven Fanless PM Sensor with Wireless Data Transmission Capabilities
[Patent] 무인 비행체용 미세먼지 측정장치 · Application number: 10-2020-0055702 · Publication number 10-2021-0137650
2016.03 – 2022.08
Full Scholarship of Samsung Semiconductor
Sungkyunkwan University
Academic Service
Reviewer, ICLR 2027
Open Source Contribution
Restored BLT's 512-byte sliding-window attention in the entropy patcher and local encoder/decoder, where the dropped config collapsed the patch rate from 4.07 to 1.15 bytes/patch past 512 bytes.
Software
Saves tmux sessions together with the Claude Code conversation running in each pane, and brings them back after a reboot — same windows, splits and directories, with each conversation resumed.