CV
General Information
Full Name | Su Wang |
Languages | English, Mandarin |
Education
-
2015-2020 Ph.D.
The University of Texas at Austin - Major | Computational Linguistics (4.0)
- Minor | Statistics (M.S., 4.0)
- Advisors | Greg Durrett, Katrin Erk
- Topics | NLP, Bayesian Methods, Neural Networks
-
2012-2015 M.A.
University at Buffalo, SUNY - Major | Linguistics (4.0)
- Minor | Computer Science (4.0)
- Advisors | Jürgen Bohnemeyer, Jean-Pierre Koenig
- Topics | Semantics, Typology
Experience
-
2023 - Present Senior Research Engineer
Google DeepMind - Multimodality (focusing on LLM, VLM). Model finetuning work.
- Evaluation (alignment, fairness, creativity). Used diagnostic evaluations to improve the objective designs for T2I models.
-
2020 - 2023 Research Engineer
Google Research - Natural Language Processing. Explored methods in intention parsing from natural languages.
- Vision-Language Navigation. Created and deployed highly scalable training data for VLN agents.
- Text-to-image Generation/Editing. Built SoTA T2I editor, along with a comprehensive benchmark.
-
2019 Research Intern
Google AI Language - Natural Language Grounding and Navigation. Contributed to VLN instruction evaluation (ECCV'21 publication).
-
2018 Research Intern
Google Assistant - Paraphrasing. Collected strong evidence for the value of semantic augmentation in paraphrase generation (in particular for improving human ratings).
-
2016 - 2017 Software Engineer
OJO Labs, Inc. - Natural Language Processing. Contributed to conversational infra.
- Machine Learning. Implemented SoTA models to assist in infra upgrades.
- Conversation Modeling. Explored conversational modeling and investigated multiple potential directions.
Tech Stack
-
Research
- Broad | NLP, CV, ML, GenAI
- Focus | Multimodality, evaluation
-
Languages & Frameworks
- Languages | Python (main), R, Prolog, C++, Java, HTML/CSS/JS
- Framework | TensorFlow, PyTorch, JAX
Research Interests
-
Multimodality
- Text-image semantic alignment
- Fine-grained / diagnostic multimodal evaluation
-
Human-Agent Interface
- Coding translation
- Universal prompt optimization
Other Interests
- Cycling (road)
- Casual (formerly competitive) gaming