Email: [email protected]
Mobile: 0955039451
Home: Sanchong Dist., New Taipei City, Taiwan
Apr 2021 - Nov 2021
Main Work:
1. The New Taipei City Government has stepped up efforts to ban littering used by object detection models.
2. The National Science and Technology Center for Disaster Reduction needs to forecast river water levels and prevent flooding via IOT sensor data.
Others:
1. Develop POS machine by Vue, and design shop system in weekend.
2. Design a shopping website, including product list, shopping cart, member registration, administrator and so on.
Feb 2020 - Apr 2021
Main Work:
1. Instead of using triangulation, use a machine learning model to determine the threshold of RSSI which is always unstable.
2. Upgrade the TF1 to TF2, and improve model accuracy in lots of cases: just mention a few-Mackay Memorial Hospital(馬偕), BMW, Leaspy(宇博), Hannstar(瀚宇彩晶) and etc.
3. Conduct experiments and write tests for new products, for beacons transmitted by BLE.
4. System maintenance and repair
Other research:
1. For the purpose of improving the accuracy of positioning, search the object detection algorithms.
2. Demo Raspberry Pi 4 with Google Coral Accelerator and run related object detection models on the device
Sep 2018 - Feb 2020
Main Work:
1. Worked on automated optical inspection of panels and light guide plates, and helped factories to reduce labor costs.
2. Used Energy-based models to achieve automatic detection finding the defect in panels.
3. Then, determined the threshold by OpenCV
4. Moreover, Semi-Supervised learning can help to define what the defect is, this ability resembles the cognition of human operators, and the concept of out-of-distribution may be a good way to determine the threshold.
5. With the light guide plates, the results of the algorithm at the beginning of the cutting tests: accuracy 0.814 (base: 0.643), F1 score: 0.691
6. Applying for a patent for the above-mentioned algorithm
Others:
1. Trained deep learning models on GPUs efficiently, and built docker files for personal projects in order to create an isolated environment which would not interfere with other people’s projects.
2. Built an automatic daily mail system for reporting the data of production from the factory and the conditions of the machinery and equipment in the factory.
2016 - 2018
Thesis: ECG Classification with Siamese Network
Sarcasm Detection through Word2vec and Convolutional Neural Network (Published on The 26th South Taiwan Statistics Conference)
On Feature Combination for Sentiment Classification (Contributed to IEEE Intelligent Systems)
ECG Classification with Convolutional Neural Networks (Accepted by 2018 GCEAS Global Conference on Engineering and Applied Science)
2012 - 2016
Leveraging Text Mining and Sentiment Analysis to Increase Vehicle Sales
My name is Tseng Po-Yen, majored in electrical and electronics engineering in National Taipei University and the department of industrial engineering and management in National Chiao Tung University. I devoted to crawling and sentiment analyzing when I was in college, and delved into deep learning in ECG classification. In addition to studies, I participated in volleyball team and sometimes did sports on the weekend.
When worked in Coretronic, I was responsible for automated optical inspection of panels and light guide plates and achieved accuracy 0.814 (F1 score 0.691). Last, the above-mentioned algorithm applied for patent. Next, I dedicated in indoor positioning using deep learning in THLight. This model can get test accuracy about 0.99 in lots of cases better than the past model (test accuracy about 0.93). Now, I design inference server and takes some cases in Mitac. Beside training model, I also try to use docker and K8s to deploy edge computer. Last, apart from work, I also joined a club Deep Learning 101, and had a speech on VQVAE.
Email: [email protected]
Mobile: 0955039451
Home: Sanchong Dist., New Taipei City, Taiwan
Apr 2021 - Nov 2021
Main Work:
1. The New Taipei City Government has stepped up efforts to ban littering used by object detection models.
2. The National Science and Technology Center for Disaster Reduction needs to forecast river water levels and prevent flooding via IOT sensor data.
Others:
1. Develop POS machine by Vue, and design shop system in weekend.
2. Design a shopping website, including product list, shopping cart, member registration, administrator and so on.
Feb 2020 - Apr 2021
Main Work:
1. Instead of using triangulation, use a machine learning model to determine the threshold of RSSI which is always unstable.
2. Upgrade the TF1 to TF2, and improve model accuracy in lots of cases: just mention a few-Mackay Memorial Hospital(馬偕), BMW, Leaspy(宇博), Hannstar(瀚宇彩晶) and etc.
3. Conduct experiments and write tests for new products, for beacons transmitted by BLE.
4. System maintenance and repair
Other research:
1. For the purpose of improving the accuracy of positioning, search the object detection algorithms.
2. Demo Raspberry Pi 4 with Google Coral Accelerator and run related object detection models on the device
Sep 2018 - Feb 2020
Main Work:
1. Worked on automated optical inspection of panels and light guide plates, and helped factories to reduce labor costs.
2. Used Energy-based models to achieve automatic detection finding the defect in panels.
3. Then, determined the threshold by OpenCV
4. Moreover, Semi-Supervised learning can help to define what the defect is, this ability resembles the cognition of human operators, and the concept of out-of-distribution may be a good way to determine the threshold.
5. With the light guide plates, the results of the algorithm at the beginning of the cutting tests: accuracy 0.814 (base: 0.643), F1 score: 0.691
6. Applying for a patent for the above-mentioned algorithm
Others:
1. Trained deep learning models on GPUs efficiently, and built docker files for personal projects in order to create an isolated environment which would not interfere with other people’s projects.
2. Built an automatic daily mail system for reporting the data of production from the factory and the conditions of the machinery and equipment in the factory.
2016 - 2018
Thesis: ECG Classification with Siamese Network
Sarcasm Detection through Word2vec and Convolutional Neural Network (Published on The 26th South Taiwan Statistics Conference)
On Feature Combination for Sentiment Classification (Contributed to IEEE Intelligent Systems)
ECG Classification with Convolutional Neural Networks (Accepted by 2018 GCEAS Global Conference on Engineering and Applied Science)
2012 - 2016
Leveraging Text Mining and Sentiment Analysis to Increase Vehicle Sales
My name is Tseng Po-Yen, majored in electrical and electronics engineering in National Taipei University and the department of industrial engineering and management in National Chiao Tung University. I devoted to crawling and sentiment analyzing when I was in college, and delved into deep learning in ECG classification. In addition to studies, I participated in volleyball team and sometimes did sports on the weekend.
When worked in Coretronic, I was responsible for automated optical inspection of panels and light guide plates and achieved accuracy 0.814 (F1 score 0.691). Last, the above-mentioned algorithm applied for patent. Next, I dedicated in indoor positioning using deep learning in THLight. This model can get test accuracy about 0.99 in lots of cases better than the past model (test accuracy about 0.93). Now, I design inference server and takes some cases in Mitac. Beside training model, I also try to use docker and K8s to deploy edge computer. Last, apart from work, I also joined a club Deep Learning 101, and had a speech on VQVAE.