Hi, I am Jinyan, currently a second year CS PhD student at Cornell university advised by Claire Cardie and Wen Sun.

Research interest: I am currently interested

• LLM reasoning: how to improve LLM to think and solve complex problem such as math and coding.
• LLM personalization: how to personalize LLM without using resource intensive approaches such as fine-tuning.
• Retrieval augmented generation: how to design better and controlable RAG system to fit for different user cases and balance factual correctness and contextual correctness. (Check out my latest work in evaluating the robustness of RAG system when poisoning rag corpus Eval-PoisonRAG and user customizable adaptive retrieval Flare-Aug)

In the past, I worked on safe and responsible use of AI/LLM, including

• Design algorithms that are tailored to diverse user preferences (see MSGD)

• Safeguard AI from misuse such as spreading misinformation, for example, I design algorithm and build benchmarks such as M4 and M4GT-Bench, and methods such as DetectLLM for detecting machine generated texts, and fake news detection.

• Design machine learning algorithms that are differentially private (See works such as Fast DPSCO, Non-interactive LDP; DPSCO in non-Euclidean space)

A complete list can can be find here

📫 Contact

Email: js3673[@]cornell.edu