Jun 2024 – Sep 2024 · 3 mos

• Trained multi-modal language models to reason and act

Dec 2022 – May 2023 · 5 mos · Menlo Park, California, United States

• optimized FB stories ranking algorithm by tuning value model weights with Multi-Objective Bayesian Optimization
• built Multi-task Multi-label models and optimized them with Mixture of Experts modules and auto gradient balancing techniques
• saved multi-feed CPU by >0.6% (considered a significant efficiency win to the company)

Aug 2022 – Nov 2022 · 3 mos · Mountain View, California, United States

• Implemented prompt-tuning techniques on top of transformers-based language models
• Finetuned foundation language model LaMDA with prompt-tuning techniques and vision-language model CoCa on the realistic screen navigation dataset MoTIF, where our models outperform STOA baselines

Dec 2020 – May 2021 · 5 mos · Beijing, China

• worked on OpenAttack, an open-source toolkit for developing adversarial attacks and defenses
• OpenAttack was accepted by ACL2021 demo track

Mar 2020 – Jun 2022 · 2 yrs 3 mos · Stanford, California, United States

• Led the research project ELIGN, which was accepted by NeurIPS 2022
• Led the research project CREPE, which was accepted by CVPR 2023 as a highlight paper (2.5% of submissions)

Jul 2019 – Jun 2022 · 2 yrs 11 mos · Shanghai City, China

• • Tutored high school students on Bioinformatics, specifically on using R, Python and Biopython for data analysis

Jun 2019 – Aug 2019 · 2 mos · Ningbo, Zhejiang, China