Biography

Jinkyu Kim is an Assistant Professor in the Department of Computer Science and Engineering at Korea University. Prior to his academic appointment, he served as a Research Scientist at Waymo (formerly Google’s self-driving car project), where he led advanced research initiatives in autonomous driving, particularly addressing critical challenges in planning and behavior prediction. He earned his Ph.D. in Computer Science from the University of California, Berkeley, under the supervision of Professor John (the edge) Canny. During his doctoral studies, he was affiliated with the Berkeley AI Research (BAIR) Lab and the Berkeley DeepDrive (BDD) project. His research focused on developing explainable and advisory models—systems capable of articulating their decision-making rationale, delineating their strengths and limitations, and providing insights into their anticipated behaviors. His current research interests include: (1) Multi-modal representation learning across image, text, and sound, (2) Scalable machine learning for autonomous vehicles, (3) Continual and life-long learning, and (4) Foundation models.

Download my resumé.

Interests
  • Multi-modal Representation Learning
  • Machine Learning for Self-driving Vehicles at Scale
  • Continual and Life-long Learning
  • Foundation Models
Education
  • PhD in CS, 2019

    Univ. of California, Berkeley

  • MS in ECE, 2010

    Korea University

  • BS in EE, 2008

    Korea University

Recent Publications

(2025). Mitigating Trade-off: Stream and Query-guided Aggregation for Efficient and Effective 3D Occupancy Prediction. In ArXiv.

PDF Cite

(2025). Querying Labeled Time Series Data with Scenario Programs. In NASA Formal Methods.

PDF Cite

(2025). 3D Occupancy Prediction with Low-Resolution Queries via Prototype-aware View Transformation. In CVPR2025.

PDF Cite

(2025). DiffExp: Efficient Exploration in Reward Fine-tuning for Text-to-Image Diffusion Models. In AAAI2025.

PDF Cite

(2024). Sparse-to-Dense LiDAR Point Generation by LiDAR-Camera Fusion for 3D Object Detection. In ArXiv.

PDF Cite

(2024). ENTP: Encoder-only Next Token Prediction. In ArXiv.

PDF Cite

(2024). InstructBooth: Instruction-following Personalized Text-to-Image Generation. In ArXiv.

PDF Cite Project

(2024). H-Direct: Homeostasis-aware Direct Spike Encoding for Deep Spiking Neural Networks. In NeurIPS Workshop 2024.

Cite

Members

Full-time

Members

Part-time

Teaching

CS Colloquium (COSE406)
Introduction to Self-driving Cars (COSE472)
Applications and Understanding of Artificial Intelligence (AICE401)
Deep Learning (AAA642)
Algorithms (CS214)
Pattern Recognition (AAA619)
Data Science (COSE471)

Contact

  • jinkyukim@korea.ac.kr
  • Dept. of Computer Science and Engineering, Korea University, Seoul, 02841
  • 605 Woojung Informatics Building
  • Thursday 3:00 to 4:00