Senior Manager  •  Wistron

七月 2021 - Present

I switched to a managerial role, managing four employees, and the original job turned into planning, task assignment, technical survey, and also included:
1. Employee performance management.
2. Employee functional development planning.
3. Hold an AI patent conference.
4. Conceive the application of MIT new technology.
5. Research MLOPS

Technical Senior Manager  •  Wistron

九月 2020 - 七月 2021

I'm currently leading a group and working on image processing and deep learning, here's a brief introduction:
1. Detecting small objects in High-resolution images: we use deep learning to identify small eggs on High-resolution images:
-This project is from a government unit used to quickly screen/count dengue eggs.
- The work content includes the design and development of image modules and back-end APIs, as well as Docker.
- 10x increase in efficiency (10 minutes vs 1 minute)
- Average error rate 15% lower than Linker (MAPE: 15% vs 25%)

2. Image identification on a fast screening reagent:
-This project comes from the first few Japanese customers who produce rapid screening reagents. The customer needs a machine to automatically identify positive or negative.
-Designed an algorithm that can identify whether the result of disease detection on the embedded computer is positive or negative.
-Because the LED color temperature and the placement of the Camera of each device will be different. A method of using software to correct the image on the device is designed.

As image processing/deep learning Group Leader, new features development, host the weekly group meeting regularly and trouble Shooting, coordinated member’s task.

MIT - Visiting Researcher   •  Wistron

二月 2020 - 九月 2020

In 2020, I was fortunate to be a Visiting Engineer at MIT for eight months, I focus on the following topics with MIT professor:
1. Unsupervised learning algorithm
-Voxel morph algorithm.
-Application of supervised and unsupervised learning in the image detection of defective components in the factory.

2. Discussion on the processing of incorrectly marked data
-Use Confidenct Learning to remove incorrectly marked data.
-Use Co-teching to reinforce training results
- This method defeated the previous algorithm, the lowest leak dropped from 1.2 to 0.7

3. Host meetings between MIT and internal members of the company.

AI/DL Engineer  •  Wistron

九月 2015 - 二月 2020

During this time, I participated in the following projects related to image processing/deep learning:
1. Liver/tumor semantic segmentation network:
- Cooperate with a hospital to develop a liver disease recognition system (has been exhibited in Nangang Exhibition Hall)
-Using a 3D semantic segmentation network to identify five different disease areas
-Processing various medical image formats (DICOM, NII)

2. As a team leader, I participated in the company's Golden Mind competition, and the baby camera project won the third place.
-Develop an image classification system on Raspberry pie.
-Learn the division of labor and cooperation on AIOT and the thinking model of start a business.

3. Improving the neural network for the face recognition system.
-The original team used openface, without major adjustments to the architecture, using fine-turning technology to allow the Network to fit smaller groups of datasets

I participated in the Golden Mind competition organized by the company as a team leader. The proposal of baby camera won the third place. I learned the AIOT division of labor and entrepreneurship mode.