Research techniques in deep learning and built question answering model with near state of the art result. Experienced in building various NLP and Computer Vision tasks. Skilled in deep learning, machine learning, statistics, problem solving, and programming. Extremely willing to learning , passionate in latest tech and quick learner.
Research and Apply Artificial Intelligence to build smart intelligence system for society.
Built Chinese sentiment and emotion analysis models by collecting, preprocessing and analysing Weibo post data, applying state-of-the-art NLP research result in model building. Achieved 40% increase in performance.
- Built automated, algorithmic agency portfolios web scraping system in Python - Facilitated client and agency matching process by building text-based image search system - Built Recommendation engine for finding similar vendors based on their skills and descriptions.
2018-04 - 2018-11
Sentiment Analysis on Stanford Treebank Dataset
- Building end-to-end neural network using improved LSTM (DC-BiLSTM), biattentive classification network and enhanced word embeddings (Contexualized word vectors).
2018-10 - 2018-11
Machine Comprehension on SQuAD Dataset
- Built an end-to-end bidirectional attention with self-attention deep neural network model to perform machine comprehension on SQuAD
- Applied Natural Language Processing (NLP) techniques such as LSTM, bi-directional attention flow, self-attention, character CNN, Pointer Network, ; Model trained using TensorFlow framework.
- Achieved near state-of-the-art result (75.9 F1 score) on dev set compared using single model.
Deep Learning Specialization | deeplearning.ai on Coursera
Developed profound knowledge of hottest AI algorithms, mastered deep learning from its foundation to its industry applications (object detection by YOLO in autonomous driving, Neural Style Transfer, Speech Recognition etc.)
2017-9 - 2018-2
Machine Learning | University of Washington on Coursera
Developed an applied understanding of machine learning models with applications such as automate calculation of credibility in loan lending