郭哲宇 / Zhe-Yu Guo
Various machine learning and deep learning algorithms are used to solve the problems of many cooperative manufacturers. The research fields include time series analysis and prediction, image recognition, natural language processing, and data reduction. The topic of the master thesis is to predict the future generation of solar power plants through machine learning methods. It can help solar power plant managers make decisions and submit papers to the International Conference on Information Management (ICIM 2020), which has been accepted.
Military Service Status: Discharged
Machine Learning Engineer, Data Scientist, Data Analyst
Taipei, Taiwan
0988622949
Skills
Deep learning
- ANN, CNN, RNN, LSTM, GRU, GAN, AutoEncoder, etc.
- Image process: YOLO, SSD, VGG, ResNet, etc.
- Natural Language Processing: Word2vec
Database
- MySQL
- InfluxDB
Website development
- Node.js
- JSP
- HTML, CSS, Bootstrap
- JavaScript
Programming
- Python：Keras, Scikit-learn, Gensim, NLTK, Numpy, Pandas, Matplotlib, Plotly, etc.
- Java
Machine learning
- Clustering: K-Means
- Classification / Regression: SVM, KNN, Random Forest, etc.
- Dimensionality reduction: PCA, MDS, etc.
Cloud platforms
- Google Cloud Platform：Cloud Speech-to-Text, Firebase
Mobile Application
- Android
Game development
- Game engine：Unity
- Image recognition: Vuforia
- C#
Hardware development
- Arduino
Education
National Yunlin University of Science and Technology, Master's degree, Information Management, Sep 2018 ~ Jun 2020
- Equipment manager of Intelligent Database Systems Lab (IDSL), responsible for the management of laboratory equipment and network.
- In the case of industry-university cooperation, through various machine learning methods to help partner companies solve problems.
- Exchange with Chiba University, Japan, and participate in professional courses related to welfare robots.
National Kaohsiung University of Science and Technology, Bachelor's degree, Information Management, Sep 2014 ~ Jun 2018
Work Experience
Hamastar Technology Co., Ltd. 哈瑪星科技股份有限公司, Engineer, intern, Jul 2017 ~ Jul 2018
- Create AR / VR games through Unity, and use Arduino to develop the equipment needed for games.
Machine Learning Projects
Solar Photovoltaic power forecasting
- Predicting the long-term and short-term generation of solar power plants can assist power plant managers in making decisions.
- Time series problem. Therefore, deep learning methods such as CNN, RNN, LSTM are used to extract weather features and predict future power generation.
- The clustering algorithm K-Means is used to segment the data of different weather to improve the short-term solar photovoltaic power forecasting performance.
- Related technologies: Keras, CNN, RNN, LSTM, Scikit-learn, K-Means
Pig location detection
- When the sow is in heat, notify the pig farm for artificial insemination. When the boar passes by, you can confirm whether it is in estrus by the movement of the sow.
- Read pre-trained models and use public pig image data sets on the Internet for transfer learning to predict sow location.
- Related technologies: Tensorflow, Object Detection, YOLO, SSD, OpenCV
Disease classification
- The patient's medical record is used to determine the type of disease the patient has.
- After cutting the text description, use NLTK to perform pre-processing such as lemmatization and removing stop words. Then train the Word2vec model to convert the words into word vectors, and predict the disease category through LSTM, HAN and other time series deep learning methods.
- Related technologies: Gensim, Word2vec, NLTK, Keras, LSTM, HAN
Signal compression and equipment anomaly detection
- Compress the vibration signal of the device to reduce the required storage space, and confirm whether the device is abnormal by the vibration signal.
- The Fourier transform is used to extract the signal period, and the AutoEncoder model is trained to extract features in the vibration signal.
- Related technologies: Keras, AutoEncoder, CNN, Fast Fourier Transform
Face detection
- After detecting all the faces in the photo through Dlib, use OpenCV to mark the positions of all the faces.
- Related technologies: OpenCV, Dlib
Mixed-type data
- Mixed datasets contain numeric and text attributes, but because text cannot perform mathematical operations, it is necessary to convert text to vectors.
- Train the Word2vec model to convert the text in the field into word vectors, and then reduce the dimensions of the data through PCA and MDS to visualize the data.
- Related technologies: Gensim, Word2vec, Scikit-learn, PCA, MDS
Unstructured data file classification
- Process text data through One-Hot Encoding and TF-IDF, and use SVM to predict document types.
- Related technologies: Scikit-learn, TF-IDF, SVM
Website Development Projects
Website development - Solar Photovoltaic power forecasting System
- The results of the solar power plant's predicted power generation are presented on the webpage for the solar power plant administrator to make a decision.
- Read the predicted power generation results of a specific solar power plant and a specific time from InfluxDB, and plot the results on a web page.
- Related technologies: Node.js, JavaScript, Bootstrap, InfluxDB
Website development - Fitness system
- After the user pulls the elastic band, the tensile force is presented on the web page and the user is given appropriate suggestions.
- After obtaining the bending value on the elastic band through Arduino, build the Node.js website to present the value and give proper feedback.
- Related technologies: Arduino, Node.js
Website development - Yuntech CHC
- Frontend: Users can view the department introduction, enrollment information and the latest news.
- Backend: The background manager can add, delete and modify posts and upload active pictures.
- Related technologies: JSP, Struts2, Hibernate, MySQL
Website development - Sensor learning
- Frontend: Users can view the latest news and download files.
- Backend: The manager can add, delete and modify the posted posts and upload files.
- Related technologies: JSP, Struts2, Hibernate, MySQL
Involve Projects
Detect syringe injection
- The user assists the injection movement through the syringe model in the VR game, so it is necessary to detect the injection progress and link it to the game.
- Detect the distance through the ultrasonic sensor and return it to the computer via WiFi, then read the sensor value in the game and synchronize the injection action.
- Related technologies: Arduino, Unity
.
