I am a Ph.D. student in Computer Science at Cornell University. I am advised by Claire Cardie, with additional mentorship from Thorsten Joachims, Kianté Brantley, Daniel D. Lee and many other amazing collaborators. Outside of Cornell, I have spent time at Microsoft Research as a research intern, working with Dipendra Misra. Prior to Cornell, I worked with Luke Zettlemoyer and Yejin Choi at the University of Washington.

My primary research interests are natural language processing and machine learning. My research develops AI systems designed for effective real-world use. Standard practices in AI assume tasks are fully understood and clearly specified from the start, which is often violated in practice. Users often find it difficult to communicate their full range of needs and expectations, while models struggle to learn from and adapt to the practical objective once deployed. Although recent advances in instruction following research partially address these issues, their training requires preference data collected from paid annotators, not enable models to continually improve post-deployment, and still dependent on users’ prompt engineering skills.

To overcome these challenges, I design LLM systems that learn from naturally occurring user feedback in deployment, and develop adaptable, reliable methods to enhance their practical utility. My recent research includes:

• Continual learning from user interaction via explicit feedback in model deployment [1, 2]
• Interactive learning for LLM agents based on implicit user edits to infer latent user preference [3]
• End-to-end optimization of LLM-based retrieval pipelines based on practical feedback to the end task [4]

[Google Scholar] [Semantic Scholar] [GitHub] [Twitter]

Check out our MiniTorch project – a Python re-implementation of the Torch API. It is designed as a teaching library to teach machine learning engineers 1) core deep learning concepts such as tensor operations, autograd, and CUDA programming, and 2) how to engineer machine learning systems to be correct, robust, and fast. We built this DIY deep learning library primarily for the Machine Learning Engineering course at Cornell Tech.

Research

	I Could’ve Asked That: Reformulating Unanswerable Questions Wenting Zhao, Ge Gao, Claire Cardie, and Sasha Rush EMNLP 2024 PDF BIBTEX CODE DATA
	Aligning LLM Agents by Learning Latent Preference from User Edits Ge Gao*, Alexey Taymanov*, Eduardo Salinas, Paul Mineiro, and Dipendra Misra NeurIPS 2024 PDF BIBTEX CODE SLIDES POSTER
	Policy-Gradient Training of Language Models for Ranking Ge Gao, Jonathan D. Chang, Claire Cardie, Kianté Brantley, and Thorsten Joachims FMDM@NeurIPS 2023 PDF BIBTEX SLIDES POSTER DATA
	Continually Improving Extractive QA via Human Feedback Ge Gao*, Hung-Ting Chen*, Yoav Artzi, and Eunsol Choi EMNLP 2023 PDF BIBTEX CODE POSTER
	Simulating Bandit Learning from User Feedback for Extractive QA Ge Gao, Eunsol Choi, and Yoav Artzi ACL 2022 & UpML@ICML 2022 PDF BIBTEX CODE SLIDES POSTER TALK
	MiniTorch: Teaching Library for Python Re-implementation of the Torch API Sasha Rush, Ge Gao, Anton Abilov, and Aaron Gokaslan 2020 WEBSITE CODE
	Neural Metaphor Detection in Context Ge Gao, Eunsol Choi, Yejin Choi, and Luke Zettlemoyer EMNLP 2018 PDF BIBTEX CODE SLIDES

* denotes equal first authorship.

Contact

ggao asperand cs dot cornell dot edu

2 West Loop Rd, Cornell Tech, NYC

Misc

• You could pronounce my first name as /gə/ or G.
• If interested, read more about my full name in classical Chinese literature.