Po Chih Huang


My name is Po-Chih Huang, aka Brian Huang
M.S. student in Computer Science at National Taiwan University
Advised by Prof. Pu-Jen Cheng
My research focus on deep learning and NLP
Currently, I am a software engineer in Microsoft Ads team

Please find my latest resume here

Publications


Target Attention Network for Targeted Sentiment Analysis, ACLCLP '18

Apply target attention on sentiment target to focus on more relavent parts
Also introduce novel position embedding which directly models the location relation between the target and its context paper link

Awards


Master Thesis Award in ACLCLP '18

My master thesis about targeted sentiment analysis
Named as the master thesis award honorable mention in ACLCLP '18

Presidential Award

Named as the winner in the 2nd semester in Senior
With GPA 4.25/4.3

Research


Sentiment Analysis

Survey SA papers in Deep Learning field
Review papers, and implement interested ones
Papers come from top conference includes but not limited to ACL / EMNLP / COLING / NIPS / ICLR

research project

Generalization of Deep Learning

Why is Deep Learning so powerful?
Why can it perform well on unseen data?
Where does the generalization ability comes from?
Research discover the blackbox of Deep Learning

research article

Projects

See more on Github


Reinforcement Learning Chatbot

300+ stars on Github
Train the chatbot with a seq2seq model first
Then use RL to get more interesting results

project link

Semantic Segmentation

200+ stars on Github
An important sub-task for automated driving
Reproduce CVPR2015 best paper. Use fully convolutional network to predict pixel-wise label

project link

Image Retrieval

300+ stars on Github
Extract features from query image
Retrieve the most similar image from database

project link

Chinese TV assistant

A task-dependent chatbot
It's Chinese-based, trained by RL
Helps you to find television program and movie

project link

Video Captioning

Reproduce a paper from ICCV2015
Extract video features from VGG net
Feed it into an encoder-decoder architecture
Then output a description about the video

project link

Seat Layout Generation

Give a room photo, generate the seat layout
So you can know where the empty seats are
This technique can expand to other scenario
Such as counting cars in a parking area

project link

Skills


Data Analyzing

Deep Learning
Machine Learning
Natural Language Processing

Languages & Tools

Python
Pytorch
Tensorflow
Keras

Full-end development

Javascript for frontend
Python for backend
Flask for web framework

Po-Chih Huang

Contact



  

latest version: 2020/09/25 00:28

UTC+8:00