3 Inseok Kong Schedule

Undergraduate Researcher · University of Seoul

Inseok Kong

Planning to pursue a Master's degree in Fall 2027. I'm open to research collaborations on trustworthy multimodal AI, VLA models, and physical AI, so feel free to drop me an email!

Inseok Kong

About

Hello! I am an Undergraduate Researcher in the Speech & Interactive AI Laboratory (SIA) at University of Seoul (UOS), advised by Prof. Heeseung Kim. I am also a senior undergraduate student at UOS, double-majoring in GeoInformatics and Artificial Intelligence (expected graduation: Aug. 2027). Previously, I worked as an Undergraduate Research Student in the Statistical Inference and Information Theory Lab (SIIT) at KAIST (advised by Prof. Junmo Kim) and the Robotics and Computer Vision Lab (RCV) at UOS (advised by Prof. Jiyoung Jung).

I am interested in making AI systems trustworthy in real-world applications. My research began with adversarial defense for 3D point clouds. Seeing adversarial attacks as only one of many threats a model faces in practice, I broadened my focus to test-time adaptation (TTA) under distribution shifts caused by sensor noise and environmental changes. Since this loss of reliability is not unique to 3D point clouds, I then turned to other modalities, and I currently study hallucination mitigation in large vision-language models (LVLMs) and audio-visual LLMs (AV-LLMs). Going forward, I aim to extend this line of research to omni-modal models, vision-language-action (VLA) models, and physical AI.

Outside the lab, I am a Formula 1 fan who follows Max Verstappen and Oracle Red Bull Racing, and I keep Formula 1 race reports on my blog.

Research Focus
  • AV-LLM hallucination mitigationNow · SIA
  • LVLM hallucination & uncertainty calibration2026 · SIIT
  • 3D point cloud test-time adaptation2025–26 · RCV
  • Adversarial defense for 3D point clouds2025–26 · RCV
  • Omni-modal, VLA models & physical AINext

News

  1. New Our work on backpropagation-free 3D test-time adaptation, SATOR, has been accepted to WACV 2027!

  2. Thrilled to announce that our paper, MAMVI, will be presented as an Oral Presentation at ICPR 2026 in Lyon, France!

  3. Our work on 3D test-time adaptation via masked multi-view point clouds, MAMVI, has been accepted to ICPR 2026!

  4. Our work on transferable 3D point cloud adversarial defense, APC, has been accepted to CVPR 2026 Findings!

Publications

  1. SATOR
    WACV 20271st Author

    SATOR: Backpropagation-Free 3D Test-Time Adaptation via Sample-Adaptive Token Routing

    Inseok Kong, Geunyoung Jung, Jiyoung Jung

    IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)Oct. 2026 · Accepted

  2. MAMVI
    ICPR 2026Oral1st Author

    MAMVI: 3D Test-Time Adaptation via Masked Multi-View Point Clouds

    Inseok Kong, Geunyoung Jung, Jiyoung Jung

    IEEE/IAPR International Conference on Pattern Recognition (ICPR)Apr. 2026 · Accepted, Jul. 2026 · Oral

  3. APC
    CVPR 2026 Findings

    APC: Transferable and Efficient Adversarial Point Counterattack for Robust 3D Point Cloud Recognition

    Geunyoung Jung, Soohong Kim, Inseok Kong, Jiyoung Jung

    IEEE/CVF Conference on Computer Vision and Pattern Recognition Findings (CVPR Findings)Mar. 2026

  4. Mitigating Feature Collapse for Test-Time Adaptation of 3D VLMs via Distribution-Guided Regularization
    Under Review

    Mitigating Feature Collapse for Test-Time Adaptation of 3D VLMs via Distribution-Guided Regularization

    Geunyoung Jung, Inseok Kong, Jiyoung Jung

    Under Review

  5. Class Topology Survives Corruption
    Under Review

    Class Topology Survives Corruption: Prior, Monitor, and Repair for 3D Continual Test-Time Adaptation

    Geunyoung Jung, Inseok Kong, Jiyoung Jung

    Under Review

  6. AEAO
    Under Review1st Author

    AEAO: Question-Adaptive Attention Budgets for Mitigating Cross-Modal Hallucinations in Audio-Visual Large Language Models

    Inseok Kong, Heeseung Kim

    Under Review

Background

Research

  1. Aug. 2026 – PresentCurrent
    Undergraduate Research Student · University of Seoul · Seoul, S. Korea
    • Advisor: Prof. Heeseung Kim
    • Research on Audio-Vision-Language Models (AVLLMs) and hallucination mitigation.
  2. Jul. 2026 – Aug. 2026
    Undergraduate Research Student · KAIST · Daejeon, S. Korea
    • Advisor: Prof. Junmo Kim
    • Research on Large Vision-Language Models (LVLMs), object hallucination mitigation, and uncertainty calibration.
  3. Jul. 2025 – Jun. 2026
    Undergraduate Research Student · University of Seoul · Seoul, S. Korea
    • Advisor: Prof. Jiyoung Jung
    • Researched 3D Test-Time Adaptation (TTA) under severe domain shifts, point cloud robustness & adversarial defense, and cross-modal alignment in 3D VLMs.

Education

  1. Mar. 2022 – Present
    University of Seoul (UOS)
    B.S. in GeoInformatics and Artificial Intelligence (Double Major) · Seoul, S. Korea
    • Cumulative GPA: 3.8 / 4.5
    • Relevant Coursework: Machine Learning, Deep Learning, Pattern Recognition, Data Structures, Algorithms.

Volunteer

  1. Mar. 2025 – Feb. 2026
    KB RaSchool, Middle School Program (KB Dream Wave 2030)
    Learning Mentor, 3rd Cohort · Samsung Sorisaem Welfare Center · Children & Future Foundation
    • Mentored middle school students in their studies over 58 sessions.
  2. Sep. 2024 – Dec. 2024
    Seoul Volunteer Center (Seoul Donghaeng)
    Math Tutor · Community Child Center · Jungnang-gu, Seoul
    • Taught math to children weekly; 31 recognized volunteer hours (10 sessions).

Honors & Awards

  1. Creativity Award (K-DS Conference Organizing Committee)2026 K-DATA Science Conference & Hackathon
  2. Excellence Award (Dean of the College of Urban Science, UOS)Seoul RISE · 1st Urban Science AI Competition & Expo (Project: SATOR)
  3. Academic Award (UOS President)2025 Seoul Big Data Campus Competition, Grand Prize
  4. Excellence Award (Director of the Engineering Education Innovation Center, UOS)UOS Capstone Design Project (Team GeoPath)
  5. Grand Prize (Seoul City Mayor)2025 Seoul Metropolitan Big Data Campus Competition (Team GeoPath)
  6. Encouragement Award (AI Innovation Convergence University Project Group)AICOSS AI Convergence Industry-Academia Hackathon
  7. Dean's Award (Dean of the College of Engineering, UOS)AICOSS Deep Learning Winter Bootcamp

Certifications

  1. ADsP (Advanced Data Analytics Semi-Professional)Korea Data Agency
  2. SQLD (SQL Developer)Korea Data Agency

Projects

Jan. 2026 – Feb. 2026

LG Aimers 8th Cohort

LLM Compression Track

Applied model compression & optimization techniques (pruning, quantization) on EXAONE LLMs.

Oct. 2025 – Jan. 2026

Optimal Autonomous Shuttle Route Planning

Capstone Design Project & Seoul Big Data Campus

Developed route optimization algorithms using mobility big data to support autonomous shuttle services for elderly citizens in Seoul.

Mar. 2025 – Jul. 2025

Sportier

UOS Entrepreneurship Club · Lead Developer & President

Developed a sports matchmaking web application using React.js with location-based matching and competitive tier rankings.

Feb. 2025

Football Big Data Analytics

Football Big Data Camp

Investigated practical applications of LLMs and RAG in football performance analytics.

Off the Track

Max Verstappen · #3 · Oracle Red Bull Racing

I follow every race weekend with Max and Oracle Red Bull Racing, and keep race reports on the blog.

F1 race reports
Max Verstappen