Profile

Ryoga Yuzawa Ryoga Yuzawa / 湯澤 凌芽

AI / ML engineer focused on computer vision and embedded AI for camera systems, with experience in resource-constrained model deployment, DNN design including quantization and layer conversion, power-aware HW-SW co-optimization, and profiling/debugging using GPIO instrumentation and oscilloscopes. After working on interface and circuit design, I moved into developing AI-based autofocus features for the Sony Alpha series. Since 2025, I have been conducting research in computer vision and edge AI at UC Berkeley through Sony's overseas study program.

Experience

UC Berkeley
Berkeley, CA, USA
Visiting Researcher — Mechanical Systems Control Lab (Berkeley AI Research)
Aug. 2025 - Present
Sony
Tokyo, Japan
AI / ML Engineer — Camera Embedded AI
Apr. 2023 - Aug. 2025
Hardware Design Engineer — Display Interface
Apr. 2020 - Mar. 2023

Education

Shinshu University
Nagano, Japan
Master of Engineering — Electrical and Computer Engineering
Apr. 2018 - Mar. 2020
Bachelor of Engineering — Electrical and Computer Engineering
Apr. 2016 - Mar. 2018

Awards

2nd Place — CVPR 2026 Workshop, IEEE Low Power Computer Vision Challenge (LPCVC), Action Recognition in Video
May 2026
Achieved 96.77% accuracy at 19 ms latency on Qualcomm Dragonwing IQ-9075 EVK with a ViT-based model using layer conversion and knowledge distillation.
1st Place — The University of Tokyo, Agile-X Project: FPGA AI Design Hackathon
Jun. 2025
Achieved 340 FPS on an end-to-end camera recognition pipeline on AMD Kria KV260, implementing CenterNet on DPU IP with hardware, software, and AI co-optimization.

Patents

Subject recognition method for cameras
Information processing device and information processing method
Nov. 2024
WO2024232254A1, pending. Assignee: Sony Group Corp.
Transmission line, wiring board, and high-frequency device & design method
May 2020
JP6693655B2, granted.

Technical Skills

Languages: C/C++, Python

Embedded / Camera: RTOS, linker script, camera pipeline

ML / Deployment: PyTorch, TensorFlow, ONNX, quantization, knowledge distillation

HW / Debug: JTAG, oscilloscope, protocol analyzer, high-speed interfaces (I2C / SPI / MIPI DSI / CSI / LVDS)

Others

TOEIC
Score: 920 — Mar. 2023
JDLA Deep Learning for Engineer
Credential ID: 455955991 — Aug. 2023