Josh Chang
Me
Josh Chang
M.S. Student in Computer Science at Columbia University
joshspchang [at] gmail.com
Another me.
I love movies and sports, especially basketball and table tennis.

Biography

I’m Sung-Ping (Josh) Chang. I am a first year graduate student in Computer Science at Columbia University. During my undergraduate, I worked with Prof. Yun-Nung (Vivian) Chen at Miulab and Prof. Zhao-Ming Gao at Center for AI and Robotics at NTU. Besides, I love hip-hop music and basketball. I was the vice captain of NTU CSIE Men basketball team in my undergrad. I am also a big Milwaukee Bucks fan!!! If you would like to learn more about me, please see my [ Resume ] or contact me at joshspchang [at] gmail.com.

Timeline & Experiences

2022 May. - 2022 Aug.
Machine Learning Intern @ Apple
Worked for Apple Machine Translation Modeling Team
• Multilingual Neural Machine Translation (MNMT)
• Researched on Knowledge Distillation approach for MNMT
2021 Sep. -
Graduate student @ Columbia University
M.S. in Computing Science
2021 July. - 2021 Aug.
Data Research Intern @ Verizon Media
Worked on E-Commerce Recommendation System.
• BERT and Word2Vec model on Recommendation System
• GPT-2 on Item Slogan Generation
• Access and manage data using Hadoop, Spark
Manager: Alex Ou
Mentor: Chia-Hsin Ting
2020 July. - 2020 Aug.
Machine Learning Intern @ Cathay Life Insurance
Worked on Fraud Detection Projects for Data Analysis Team.
• Graph Features to find relationships between clients and salesmen
• Desicion Tree Methods on fraud detection
• Neo4j Graph Database
Mentor: Yu-Kai Huang
2020 Feb. - 2020 June
Research Assistant @ NTU AI & Robotics Center
Center for Artificial Intelligence and Advanced Robotics
Focused on building interactive system on Covid-19 news.
• Statistical / Neural Machine Translation
• Elastic Search / Google Search API
Supervisor: Prof. Zhao-Ming Gao
2019 July - 2019 Aug.
AI Engineering Intern @ IBM
Implemented and improved computer vision models.
• MobileNet-based models for Image Classification
• Knowledge Distillation methods to shink the model size
Manager: Jerry Wang
2018 Sep. - 2020 June
Undergraduate Researcher @ NTU Miulab
Machince Intelligence & Understanding Laboratory
Researched on different NLP problems:
• Dialogue Systems
• Natural Language Generation
• Dialouge State Tracking
Supervisor: Prof. Yun-Nung (Vivian) Chen
2016 Sep. - 2021 Jan.
Undergraduate student @ NTU
B.Sc. in Computing Science
1997
He was Born

Publications

Leveraging Speaker Profile and Knowledge Enrichment for End-to-End Advising Reponse Rankings
Retrieval-based dialogue modeling has drawn a lot of attention due to its practical usage. This paper focuses on using BERT-based model to extract speaker profile information and enriching the entity-related knowledge in dialogues to better predict the next response given the dialogue contexts.
Sung-Ping (Josh) Chang, Yun-Nung Chen, Ting-Rui Chiang, Chao-Wei Huang
DSTC8 @ AAAI 2020 (Poster) New York, USA
[ pdf ] [ poster ]

Honors & Awards

    Machine Learning / Deep Learning

  • Silver Medal (Ranked 21th out of 852 teams) / MLB Player Digital Engagement Forecasting / 2021
  • 1st Prize with TWD $200,000 (out of 469 teams) / AICup – Thesis Tagging Competition / 2019 [ code ]
  • 1st Prize with TWD $160,000 (out of 244 teams) / Cathay Big Data Competition / 2019 [ slides ]
  • 4th place with Intel Neural Compute Stick 2 (out of 148 teams) / AICup Course Challenge / 2021 [ code ]
  • 8th place wtih TWD $10,000 (out of 421 teams) / AICup – Vocal Transcription Competition / 2020 [ code ]
  • Ranked 31th (out of 823 teams) / Shopee Code League - Product Detection / 2020 [ code ]
  • Honorable Mention with TWD $5,000 / NTU CSIE Special Topic Research Exhibition / 2019

Selected Projects

Graph Neural Networks(GNN) in Tensorflow
The project aims to implement state-of-the-art GNN models. It have already contributed to Neural Structure Learning, Tensorflow. There are three GNN models: Graph Convolution Network(GCN), Graph Attention Network(GAT), Graph Isomorphism Network(GIN) in the repository now. It also supports the sparse tensor version of GNN models to increase efficieny and accuracy.
[ code ]
Fine-tuned BERT Encoder & GNN for Node Classification
BERT and GNN are the most popular neural networks models in NLP and graph dataset. Therefore, we want to bring the advantages in BERT to the citation network, and evaluate that how much improvement can BERT help GNN models work. We implemented our method on ogbn-arxiv dataset and the result beats the 1st place in Open Graph Benchmark Leaderboard!
[ code ]
PTT Gossip Browser (2020 Spring IR Final Project)
Everyone loves gossip. But, the information in PTT Gossiping is quite messy. The project aims to to build a retrieval system for news in PTT Gossiping by Vector Space Model(VSM). We also implemented the recommendation system and website interface to be more convenience for potential users.
Josh Chang, Yu-Chun Chen, Jen-Chieh Yang, Zi-Yuan Huang
[ code ] [ report ] [ demo ]
Smart Outfit (2019 Spring HCI Final Project)
Choosing what to wear is difficult for everyone. The project focused on how to design a interface which can recommend suitable outfits for potential users. We developed an ios app to record users’ preferred clothes and recommend users their best outfits based on the weather information and the latest trend.
Josh Chang, Yu-Chun Chen, Can Lu
[ website ] [ poster ] [ demo ]
Top