Jinsong Zhang

Master Student @ UCSB
Email: jinsongzhang[at]ucsb[dot]edu

I am currently pursuing a M.S. degree in Computer Engineering at the University of California, Santa Barbara. I previously earned my B.Eng. in Electronic Engineering from the University of Liverpool.

My current research, supervised by Prof. Zheng Zhang, focuses on low-precision tensorized DNN training and hardware acceleration. Previously, I worked in the WINDY Lab under Prof. Shiyu Zhao on autonomous systems.

Outside of academics, I am passionate about envisioning the future of computing architectures, from brain-inspired computing to potential bio-integrations. I also enjoy playing basketball and exploring new landscapes. For my recent work and reflections, you can find them here: Reflections.

News

Education

Skills

Experience

Selected Research Projects

Tensorized NN
Comprehensive Design Space Exploration for Tensorized Neural Network Hardware Accelerators

Jinsong Zhang*, Minghe Li*, Jiayi Tian, Jinming Lu, Zheng Zhang

Submitted to 2026 Design Automation Conference (DAC)

Proposed a novel hardware accelerator architecture for low-precision tensorized neural networks. We performed a comprehensive design space exploration on FPGA, achieving significant improvements in latency and energy efficiency compared to baseline implementations.

YOLOv5
Visual Perception for Small Cars Based on YOLOv5s via ROS & Triton

Jinsong Zhang, Jiachen Liang, Zhao Ma

Deployed a YOLOv5s object detection model on an embedded platform using the NVIDIA Triton Inference Server. Integrated the perception pipeline with ROS for real-time autonomous navigation of small-scale vehicles.

RNN
Aircraft Trajectory Prediction using LSTM & GRU

Jinsong Zhang

Developed and compared Recurrent Neural Network models (LSTM and GRU) to predict aircraft flight trajectories based on historical ADS-B data. Analyzed prediction accuracy and latency trade-offs.

Tucker Tensor
Tucker Tensor Layer in Fully Connected Neural Networks

Jinsong Zhang

Implemented Tucker decomposition for fully connected layers to reduce model parameter size while maintaining accuracy.

Selected Engineering Projects

ResNet Control
ResNet Based Visual Perception and PID Control in Autonomous Vehicle

Combined deep learning based perception with classical PID control for lane keeping and navigation.

RISC-V
Full Custom 32-bit 5-stage Pipeline RISC-V Processor

Designed a fully functional 5-stage pipeline processor supporting RV32I instruction set, handling hazards and forwarding.

Multi-sensor Slant Perception for UAVs

Integrated laser sensors and IMU data to estimate terrain slope for UAV landing maneuvers.

MIPS
FPGA Implementation of MIPS Processor

Implemented a MIPS processor on FPGA using Verilog, extending instruction support.

Flag Counter