Projects
Debias in Pre-trained Language models
- 2022 - Now
- Intro: Bias and stereotypes in large models increase the risk of NLP applications, and we want to tackle it from a new viewpoint - multitasks with the Pareto Frontier. Read more
Understanding relation prediction in Auto-Driving
- 2022 - Now
- Intro: The deadlock or crash of cars cannot be completely solved by current relation prediction module. We detect where it happens and why, with improvement on it. Read more
Facilities To defend Against Adversarial Attack
Understanding Rank Collapse in Offline Reinforcement Learning
- 2021 - Now
- Intro: Understand the current rank collapse theory in offline RL and try to design a new theory that better aligns with reality. Read more
Online robust control in smart grid
- 2019 - 2020
- Intro: Design an online robust control algorithm for smart grids. The algorithm reduces the cost of the smart grid or the pollution during the energy generation with theoretical guarantee by using a battery. Read more
Online MDP with Nonstationary Environment
- 2019
- Intro: At Harvard, I design a parameter-free algorithm for non-stationary online MDP with full-information feedback and achieve optimal dynamic regret. Read more
Big Data
- 2018
- Intro: At Google, I process big data, design new criteria to distinguish legal Chinese websites, and try to improve the search quality for Chinese websites. Read more
