Experiences
Applied Scientist at Amazon Prime Video (June 2021 - Now)
Machine Learning Engineer/Data Scientist at Docomo Innovations (June 2019 - June 2021)
Research and implement state-of-the-art ML/DL solutions and present them to clients.
Computer Vision: 1) Object detection with event camera 2) 3D style transfer with differentiable renderer 3) Crack image data augmentation with StyleGAN2 3) 3D pose extraction from 2D video
Natural Language: 1) Extract relations from free text and build knowledge graph from documents 2) Clustering and labelling relations via sentence embeddings 3) Keyword/season based Haiku poem generation with ranking system using LSTM models
Tabular Data: 1) Use LSTM-based GAN model to learn trajectory embeddings and predict customer behaviors 2) Product demand prediction with MLP model
UI Tools: 1) Build UI for data collection with Express.js and PostgreSQL 2) Build demo using Django and deploy on AWS
Progressive Domain Adaptation for Object Detection (Apr. 2018 - Dec. 2018)
Unsupervised domain adaptation to align the labeled source domain with the unlabelled target domain via adversarial learning
Utilize synthetic images as an intermediate domain to reduce the effort in aligning source and target domains
Intern at Verisk (May 2018 - Aug 2018)
Investigate and implement state-of-the-art text detection and recognition (TDR) via deep learning methods
Improve TDR on real-world document image/PDF files
Learning from Photoshop Tutorial Videos (Oct 2017 - July 2018)
Recognize and parse each command performed in Photoshop tutorial videos
Apply 3D CNN for classifying different commands
Learning to tell signal lights for autonomous driving (Sept 2016 - Sept 2017)
Investigated a learning-based method for classifying rear car light signals
Utilized and implemented an action recognition architecture which combines convolution neural networks (CNN) and long short term memory (LSTM) to classify time series image data
Intern at Institute for Information Industry, Taiwan (June 2017 - Aug 2017)
Augmented reality on smart glasses
Create point cloud using ORB-SLAM2 for tracking camera position, and implemented the Extended Kalman Filter to refine the trajectory
Teaching Assistant
CSE031 Computer Organization (Spring 2018)
CSE030 Data Structures (Fall 2017)
CSE015 Discrete Mathematics (Spring 2017)