Yekun Chai

As a machine learning researcher working on deep learning, natural language processing (NLP) and deep reinforcement learning (RL), I have broad interests in language understanding and generation, deep generative models, representation learning, multimodal seq2seq applications, etc.

I am actively seeking for Ph.D. opportunities and open to collaborate. Please check my technical blogs or drop me an email.


Education

2017-2018
The University of Edinburgh
MSc, Artificial Intelligence (with Distinction)
Thesis title: Discovering spelling variants on Urban Dictionary
Supervisor: Dr. Adam Lopez , Naomi Saphra

2013-2017
Beijing Information Sci & Tech University
BSc, Computer Auditing (Data Science track)

Publications

Yekun Chai, Shuo Jin, Xinwen Hou. Highway Transformer: Self-Gating Enhanced Self-Attentive Networks. In submission 2019

Yekun Chai, Naomi Saphra, Adam Lopez. How to Evaluate Word Representations of Informal Domain?. arXiv 2019. [code]


Industrial Experience

Machine Learning Research Associate
Mar. 2019 - present
@Institute of Automation, Chinese Academy of Sciences (CASIA)

Machine Learning Research Engineer (NLP)
Nov. 2018 - Mar. 2019
@JD AI Research of JD.com, Inc.


Projects

Playing Real-time Strategy games with Deep Reinforcement Learning
@CASIA (present)
Reproduced a deep RL agent like Quake III Arena Capture the Flag

Mastering Turn-based Strategy games with Deep Reinforcement Learning
@CASIA
Developed a AlphaStar-like neural architectures, including ResNet, Transformer, Pointer Net, Transposed CNNs, CoreLSTM modules and scatter connections.

Natural Language Understanding in Task-oriented Spoken Dialogue Systems
@ JDAI

Discovering spelling variants on Urban Dictionary
@ UoE MSc thesis [pdf]

Probing word-level and character-level representation on sentiment analysis
@ UoE MLP project [pdf]

Awards

Honorable Mention of Mathematical Contest in Modeling (MCM), 2015

rss facebook twitter github youtube mail spotify lastfm instagram linkedin google google-plus pinterest medium vimeo stackoverflow reddit quora quora