CV

You can download a PDF version of my CV here.

Education

• M.S. in Computer Science and Engineering, Yonsei University, Mar. 2024 – Feb. 2026
  • Mobile Embedded Systems Lab lab homepage.
  • GPA: 4.27/4.30
• B.S. in Computer Science, University of Seoul, Mar. 2017 – Feb. 2024
  • GPA: 3.85/4.50, Major GPA: 4.10/4.50

Experience

• Research Assistant, Jan. 2024 – Present
  • Mobile Embedded Systems Lab., Yonsei University, Seoul, Republic of Korea
  • Efficient On-device AI: Worked on advancing efficient on-device AI, exploring both model-level and system-level approaches to bridge the gap between heavy AI workloads and the limited resources of mobile devices.
  • Immersive Computing: Investigated system designs that support next-generation immersive applications including augmented reality and volumetric media.

Publications

• Conference viNPU: Optimizing Vision Transformer Inference on Mobile NPUs

  Jeho Lee, Gunjoong Kim, Chanyoung Jung, Jaehee Kim, Seonghoo Park, Hojung Cha

  European Conference on Computer Systems (EuroSys), 2026

  [Paper]

  Mobile Systems On-device AI NPU Quantization Graph Rewriting

• Article 3DGS on Your Phone: Towards Fully Immersive Mobile Volumetric Video Streaming

  Gunjoong Kim*, Seonghoon Park*, Jeho Lee, Chanyoung Jung, Hyungchol Jun, and Hojung Cha

  ACM GetMobile, 2026

  [Paper]

  Mobile Systems Immersive Computing 3D Gaussian Splatting Volumetric Video Streaming

• Article VFM in Your Hands: Optimizing Real-Time Scene Understanding for Mobile Augmented Reality

  Jeho Lee, Chanyoung Jung, Gunjoong Kim, Jiwon Kim, Seonghoo Park, Hojung Cha

  ACM GetMobile, 2026

  [Paper]

  Mobile Systems On-device AI Heterogeneous Processor

• Conference Vega: Fully Immersive Volumetric Video Streaming with 3D Gaussian Splatting

  Gunjoong Kim*, Seonghoon Park*, Jeho Lee, Chanyoung Jung, Hyungchul Jun, Hojung Cha (*Co-first authors)

  ACM Annual International Conference on Mobile Computing and Networking (MobiCom), 2025

  [Paper]

  Mobile Systems Immersive Computing 3D Gaussian Splatting Volumetric Video Streaming

• Conference ARIA: Optimizing Vision Foundation Model Inference on Heterogeneous Mobile Processors for Augmented Reality 🏆 Best Paper Award

  Chanyoung Jung*, Jeho Lee*, Gunjoong Kim, Jiwon Kim, Seonghoon Park, Hojung Cha (*Co-first authors)

  ACM Annual International Conference on Mobile Systems, Applications, and Services (MobiSys), 2025

  [Paper]

  Mobile Systems On-device AI Heterogeneous Processor

Projects

• viNPU: On-device Inference Optimization Framework (2025)
  • Built a framework to run Vision Transformers efficiently on mobile NPUs by combining sensitivity-guided mixed precision with graph rewrites. Achieved average 11.5x faster inference, up to 44.9x lower energy, with near-no accuracy loss vs FP16.
  • Tools: Qualcomm Neural Processing SDK (QNN), Aimet, PyTorch, ONNX, Android
• Vega: Mobile Volumetric Video Streaming System with 3DGS (2024–2025)
  • Designed the first system to achieve 30 FPS full-scene volumetric video streaming on mobile devices with 3D Gaussian Splatting, reducing data size by 99% while maintaining high visual quality.
  • Tools: PyTorch, OpenGL ES, TensorFlow Lite, Qualcomm Neural Processing SDK, Android
• ARIA: Heterogeneous Processor Acceleration for VFM Inference (2024)
  • Developed a system for real-time VFM inference on mobile devices by combining GPU full-frame predictions with NPU dynamic-region updates, achieving 99.9% deadline success and up to 50% accuracy improvement.
  • Tools: Qualcomm AI Engine Direct SDK (QNN), LiteRT, ONNX, Android
• Cosmos: Mobile 360 Video Streaming with Content-Aware Super-Resolution (2024)
  • Developed a content-aware SR system for real-time 360 video streaming on mobile, sustaining 30 FPS and improving tile quality by up to 21.8%p and PSNR by 3.0 dB over prior work.
  • Tools: PyTorch, TensorFlow Lite, OpenCV, OpenGL ES, Android

Skills

• Languages: C++, C, Java, Python
• ML Frameworks: PyTorch, ONNX, LiteRT (TensorFlow Lite)
• GPU/NPU Frameworks: CUDA, OpenCL, OpenGL, Qualcomm AI Engine Direct (QNN) SDK
• Platforms: Android, Linux