KiYoon Yoo

profile.jpg

[Google Scholar] [CV]

I work on post-training foundation models and vision-language-action (VLA) models for autonomous driving at Zoox. Previously, I developed real-time embodied companion agents for PUBG at Krafton and contributed to Character Chat, a production RAG-based chatbot at Naver Webtoon.

I recieved my Ph.D. from Seoul National University under Nojun Kwak’s guidance with research emphasizing the safety and robustness of language models, including areas such as adversarial attack and defense and watermarking.

news

Apr 2026 I will be joining Zoox in the Bay Area! At Zoox, I will be building multimodal LLMs to enhance environmental understanding in self-driving cars.
Feb 2025 Our paper “Nearly Zero-Cost Protection Against Mimicry by Personalized Diffusion Models” has been accepted to CVPR 2025.
Jan 2025 Our co-playable gaming agent PUBG Ally developed in collaboration with NVDIA was showcased in CES 2025 Las Vegas! (Check out the videos in NVIDIA ACE, Live Demo)
Aug 2024 I joined Krafton AI as an applied research scientist.

Work Experience

Machine Learning Engineer, Zoox
May 2026 – Present
  • Post-training foundation models and vision-language-action (VLA) models for autonomous driving.
Applied Research Scientist, Krafton
Aug. 2024 – Apr. 2026
  • Developed PUBG ALLY, a real-time embodied companion agent for in-game voice chat dialogue and strategic planning, serviced as a beta test.
  • Trained Small Language Models (SLMs) using closed-loop human feedback and synthetic data for real-time, low-latency inference on consumer GPUs.
Research Intern, Naver Webtoon
Aug. 2023 – Dec. 2023
  • Contributed to Character Chat, a production RAG-based AI chatbot serving 6M+ users.
  • Conducted research in copyright protection (1 paper at NAACL 2024 and 1 at CVPR 2025).

Selected Publications

  1. Nearly Zero-Cost Protection Against Mimicry by Personalized Diffusion Models
    Namhyuk Ahn, KiYoon Yoo, Wonhyuk Ahn, Daesik Kim, Seung-Hun Nam
    In CVPR, 2025
  2. Advancing beyond identification: Multi-bit watermark for large language models
    KiYoon Yoo, Wonhyuk Ahn, Nojun Kwak
    In NAACL, 2024
    arXiv Github
  3. Exploring Causal Mechanisms for Machine Text Detection Methods
    KiYoon Yoo, Wonhyuk Ahn, Yeji Song, Nojun Kwak
    In Workshop on Trustworthy Natural Language Processing at NAACL, 2024
  4. TEAM MIPAL at MEDIQA-M3G 2024: Large VQA Models for Dermatological Diagnosis
    Hyeonjin Kim, Min Kyu Kim, Jae Won Jang, KiYoon Yoo, Nojun Kwak
    In Workshop on ClinicalNLP at NAACL, 2024
  5. Who Leaked My Document? Robust Natural Language Watermarking through Invariant Features
    KiYoon Yoo, WonHyuk Ahn, Jiho Jang, Nojun Kwak
    In ACL, 2023
  6. Self-Distilled Self-Supervised Representation Learning
    Jiho Jang, KiYoon Yoo, Seonhoon Kim, Kong Chaerin, Jangho Kim, Nojun Kwak
    In WACV, 2023
  7. Backdoor Attacks in Federated Learning by Rare Embeddings and Gradient Ensembling
    KiYoon Yoo, Nojun Kwak
    In EMNLP, 2022
  8. Detection of Word Adversarial Examples in Text Classification: Benchmark and Baseline via Robust Density Estimation
    KiYoon Yoo, Jangho Kim, Jiho Jang, Nojun Kwak
    In Findings of ACL, 2022
  9. Position-based Scaled Gradient for Model Quantization and Sparse Training
    Jangho Kim, KiYoon Yoo, Nojun Kwak
    In NeurIPS, 2020