Zhenyu Lin
Seeking for Software Development Engineer
About Me
Hello! I'm Zhenyu Lin. I am currently a Research Assistant at SFSU Mobile and Intelligent Computing Lab, where I conduct research on efficient deep learning algorithms for resource-constrained devices. In one of my projects, namely Real-Time Machine Learning for Ultra Low-power Microcontroller, I implemented model compression techniques, achieving over 85% compression and enabling real-time processing on low-power microcontrollers. Additionally, I have also mentored high school students in an NSF-funded summer program, focusing on efficient deep learning algorithms. Besides research projects, I have also gained experience in web development. For instance, in the Full Stack University Student Center project, I developed the front-end logic using ReactJS and implemented RESTful APIs for database operations.
Bio
Work Experience
- Conducting research on efficient DL algorithms for resource-constraint devices.
- Established secure remote access through SSH and maintained network security protocols for the Linux server.
- Mentored high school students in an NSF-funded summer program, focusing on efficient DL algorithms,
- Developed a Real-time Bionic Arm Control project that won a Grand Price out of 49 projects.
Projects
Full Stack / ReactJS / NodeJS
Full Stack University Student Center
- Developed the front-end logic using ReactJS and implemented RESTful APIs to handle database CRUD operations on the back end using the NodeJS framework.
- Utilized Nginx server to efficiently handle incoming requests
- Implemented Github Action as a CI pipeline to streamline the code review process and workflow
Pytorch / Tensorflow / C / Electrical Engineering
Real-time Bionic Arm Control Via CNN-based EMG Recognition
- Developed a Convolutional Neural Network-based sEMG gesture intent recognition system for prosthetic arm control using a microcontroller.
- Integrated strategies such as transfer learning, quantization, and parameter optimization to achieve real-time performance during on-device deployment.
- Experiment results indicate high accuracy in identifying user's motor intents, serving as a contribution to future AI deployments for low-cost biomedical equipment.
Pytorch / Tensorflow / C / Electrical Engineering
Efficient Deployment Of Deep Learning Model On Cortex-M Based Microcontrollers Using Deep Compression
- Implemented L1norm pruning algorithm to compress the deep learning model
- Implemented Linear Quantization to optimize the model to 8-bit for deployment
- Achieved over 85% model compression, enabling real-time processing in 500ms in cortex M base processor
Pytorch / Matlab / Python
Toward Robust High-Density EMG Pattern Recognition using Generative Adversarial Network and Convolutional Neural Network
- Developed the RoHDE framework, utilizing a Generative Adversarial Network to generate synthetic HD EMG signals that simulate unreliable recording conditions.
- Improved gesture recognition accuracy by up to 35% for CNN-based models affected by contact artifact and loose contact disturbances.
- Introduced first solution to the robustness issue in deep learning-based HD EMG PR
Pytorch / JetsonNano / TensorRT
ExoGlovesCVFusion
- Developed a computer vision system integrated with a sensor-fusion system to control a soft rehabilitation robotic glove.
- Utilized object detection and distance estimation for the glove to move towards nearby objects and pick them up.
- Integrated the system with EMG sensors, providing a simple calibration process and minimal computational delay for upper limb prosthesis control.