Projects
Hyper-parameter optimization service
I’ve been maintaining and developing the hyper-parameter optimization service (Dr.Opt) of NCTU CS Edge-AI lab since 2019.
Dr.Opt is an ML model optimization platform consisting of:
- Hyper-parameter optimization service
- Client Python package for service connoction & project control
- Project visualization & analysis via WebUI
Client Python Package
Server
The Dr.Opt server source code is not released to the public. I am involved in front-end, back-end, database, and deployment (by Docker compose).
Deep Learning - NEU Surface Defect by YOLOv3
Build an object detection model for NEU surface defect. [Project Report] [Github]
We applied data augmentation, anchor boxes adjusting (via clustering), loss function enhancement (GIOU) and TensorRT deployment to our projects.
Big data analytics & visualization
A project about “Taiwan fruit and vegetable retail prices”, which includes building price prediction models and data visualization.
Prediction Model
Implement the time-series prediction model by prophet and generate statistical charts by plotly. [Github repo]
| Wax Apple Price Prediction | Wax Apple Seasonality |
|---|---|
Visualization
Visualize the yearly retail prices, seasonality, origin, and the relevance to weather.
Implement by d3.js. [Github repo]
| Analysis | Weather |
|---|---|
MCTS based Chinese Chess AI on a Raspberry PI cluster
It is a term project of the “parallel programming” course. We implemented the root parallelization of MCTS and deployed it on a Raspberry PI cluster (4 instances). We as well simulated our parallelization via docker-compose.
See the reference for more details: [Project Report] [Slide] [Github]
Terminal-based Course Timetable System
Implement a course selecting system by shell scripts (Bourne shell).
| Interface | Adding course |
|---|---|
 |
 |