|
I am a computer science doctoral student at The Ohio State University advised by Yu Su. My research focus is on multi-modality, specifically on embodied AI and multimodal foundation models. I received BS Cum Laude in Computer Science from University of Notre Dame in May 2020. In Notre Dame, I was a part of Notre Dame Natural Language Processing Group. Email / Twitter / LinkedIn / Google Scholar |
|
|
|
|
Currently my interests are embodied AI and multimodal foundation models. I am also interested in AI for science and multimodal knowledge base. |
|
|
|
|
|
Chan Hee Song, Jiaman Wu, Clayton Washington, Brian M. Sadler, Wei-Lun Chao, Yu Su [paper] ICCV 2023 [short version] Embodied AI Workshop at CVPR 2023 [website] |
|
|
|
|
|
|
|
Adobe Research
- I worked on an video editing assistant, with emphasize on open domain code generation problem with large language models. Worked with Gang Wu, Haoliang Wang, and Uttaran Bhattacharya. |
|
Human Language Technology Center For Excellence, Johns Hopkins University
- I implemented various state-of-the-art neural architectures and adapted to named entity recognition task. Introduced a lexical feature generated from gazetteers to the named entity recognition (NER) system. As an independent project, developed a novel named entity recognition system achieving state-of-the-art F1 score on Chinese datasets using glyph features generated from convolutional neural network. |
|
Kyndi, San Mateo CA
- I migrated in-memory graph to external database by building a GraphQL server to interact with the external database using Apollo and Typescript. Also improved accuracy and representation of the knowledge graph by writing unique graph operations on a large knowledge graph. |
|
|
 |
Used Chinese glyph information to augment neural named entity system. Introduced a convolutional neural network (CNN) architecture that achieves a state-of-art performance on Chinese Weibo NER dataset and sets a new baseline F1 score on Chinese OntoNotes v5.0 dataset. Easy to implement, train and uses less data. Robust to non-Chinese characters in the dataset. |
 |
Introduced a simple method to generate gazetteers from the Web. Generated gazetteer is used as a feature to augment neural named entity recognition system, showing improvement over baseline. Developed a method to create new training data using gazetteer entity replacement. Presented new Russian NER corpus gathered from Reddit. |
 |
Independent undergraduate researcher under supervision of professor David Chiang. I investigated the impact of sentence representation on neural machine translation to develop an optimal encoding mechanism that has a better trade-off between time and accuracy. I am using C, Python and PyTorch library to build encoder/decoder, modify neural machine translation system, and implement new equations. |
 |
Research assistant to the project which aims to find a novel method to process Vatican Secret Archives to transfer early Medieval Latin handwritten papers into a computer readable format. I analyzed various character recognition algorithm and used Python to build an auto analyzer tool for Optical Character Recognition systems |
|
|
|
Undergraduate Teaching Assistant, Theory of Computing Spring 2020
|
|
|