I am a final-year Computer Engineering Ph.D. candidate at CEI Center, Duke University. I also serve as the principal investigator (PI) for Project NAIRR240270, managing $50K in research funding from the National Artificial Intelligence Research Resource Pilot (NAIRR). My name Jianyi pronounces as “jyen” [falling tone] “ee” [flat tone] :)

I am fortunate to be co-advised by Professor Yiran Chen and Professor Hai ‘Helen’ Li. My research lies at the probabilistic modeling for generative AI and trustworthy AI, especially for Large (Vision-)Language Models and Diffusion Models. My research projects include LLM controlled decoding, LLM factuality/privacy/security, Retrieval-augmented generation for LLM, Text-rich image generation with Diffusion Model, LLM (federated) alignment, Efficient inference for LLM and Diffusion models.

Previously, I received my BS in Mathematics at Fudan University. During my undergraduate research, I had the honor to be guided by Professor Changyou Chen and Dr. Chunyuan Li working on Bayesian learning algorithms.

I am actively looking for opportunities on the industry job market!
Feel free to reach out if you would like to share opportunities, collaborate, or just chat :) You may see more details about my research here. Thank you!

News

• [May. 2025] Two papers, CoreMatching (unifying token pruning and neuron pruning for vision-language models) and SADA (accelerating diffusion models), are accepted to ICML 2025!
• [Mar. 2025] I have successfully defended my Ph.D. dissertation, titled "Advancing Deep Learning through Probability Engineering: A Pragmatic Paradigm for Modern AI". Grateful to my advisors and committee for their guidance throughout this journey.
• [Feb. 2025] Our H-CoT is reported by the Register . Many thanks to Mr. Thomas Claburn for writing about our findings. (Link)
• [Feb. 2025] Our technical report H-CoT: Hijacking the Chain-of-Thought Safety Reasoning Mechanism to Jailbreak Large Reasoning Models, Including OpenAI o1/o3, DeepSeek-R1, and Gemini 2.0 Flash Thinking is available online now.
• [Jan. 2025] Three papers, Sufficient Context (RAG for LLM), Mink++ (LLM training data detection) and PPA (Privacy protection for LLM), are accepted to ICLR 2025.
• [Jan. 2025] Our ARTIST (text-rich image generation with diffusion models) paper is selected for oral presentation at WACV 2025.
• [Nov. 2024] Our Neurips 2024 paper, SLED is now available online. The project website and code repo are launched.
• [Apr. 2024] Our newest paper on RAG, Sufficient Context: A New Lens on Retrieval Augmented Generation Systems is available online now.
• [Oct. 2024] Two papers, MLLM-LLaVA-FL (large vision-language models and FL) and ARTIST (text-rich image generation with diffusion models), are accepted to WACV2025.
• [Oct. 2024] Our CoreInfer paper, an efficient LLM inference framework, is available online now.
• [Oct. 2024] Our project on efficient inference for diffusion models, NAIRR240270, received $50K research funding. I will serve as the principal investigator in the following year.
• [Sep. 2024] Our paper, SLED: Self Logits Evolution Decoding for Improving Factuality in Large Language Models, was accepted to NeurIPS 2024. See you in Vancouver!
• [Sep. 2024] Our survey on the synergy between LLMs and FL is available online.
• [Jul. 2024] Our paper, Unlocking the Potential of Federated Learning: The Symphony of Dataset Distillation via Deep Generative Latents, was accepted to ECCV 2024.
• [May. 2024] I will spend this summer working on LLM factuality and RAG at Google Research.
• [Apr. 2024] Our Min-k%++ paper, a cutting-edge technique that accurately identifies LLM pretraining data, is open-sourced now.
• [Dec. 2023] Our FederatedGPT paper is accepted to ICASSP 2024.
• [Nov. 2023] Our FederatedGPT paper is selected for oral presentation at FLFM@NeurIPS23.
• [Sep. 2023] I will be conducting research internship at Google Research, focusing on the factuality of LLMs.
• [Aug. 2023] Our grant proposal on LLM privacy protection is accepted by Accenture.
• [Jun. 2023] I will spend the summer at Adobe Research as a research scientist intern, working on improving the text-rich image generation for diffusion models.
• [Jun. 2023] Our grant proposal on using LLMs for Verilog generation is accepted by Pittsburgh Supercomputing Center. I will serve as the principal investigator in the following year.
• [May. 2023] Our ReAugKD paper is selected for oral presentation at ACL 2023. See you in Toronto!
• [May. 2023] Our FederatedGPT paper, the first exploration of FL-based instruction tuning for LLMs, is open-sourced now.
• Show more
[May. 2023] Our ReAugKD paper, leveraging retrieval techniques to enhance the knowledge distillatin of pretrained language models, is accepted ACL 2023.
[Apr. 2023] Our paper Fed-CBS , which boosts FL performance via efficient client sampling, is accepted to ICML 2023! See you in Hawaii!
[Sep. 2022] We are invited to contribute our vision paper to the 2022 IEEE 8th International Conference on Collaboration and Internet Computing (CIC).
[Sep. 2022] Our paper, Why do we need large batch sizes in contrastive learning? A gradient-bias perspective , is accepted to NeurIPS 2022! See you in NOLA!
[Jun. 2022] I will spend this summer working on retrieval techniques and knowledge distillation for pretrained language models at Amazon Science.
[May. 2022] I pass my preliminary exam and have become a Ph.D. candidate! 🎆
[May. 2021] Our paper, FLOP, focusing on FL and Covid-19, has been accepted to KDD 2021!
[Jun. 2020] Our paper, VR-SPOS has been accepted to ICML 2020!
[Jan. 2020] Our paper, SPOS is accepted to AISTATS 2020!
[Dec. 2019] Our paper, cSG-MCMC has been selected as oral presentation at ICLR 2020!

Experience

Contact

Address: Wilkinson Building, 534 Research Dr, Durham, NC 27705
Office Location: Room 431
Email: jianyi.zhang@duke.edu