Yu Fei (@ Amazon Rufus)
@Walter_Fei
PhD student @UCIrvine working on NLP/ML. Previously: ms @ETH, undergrad @PKU1898
🗒️Can we meta-learn test-time learning to solve long-context reasoning? Our latest work, PERK, learns to encode long contexts through gradient updates to a memory scratchpad at test time, achieving long-context reasoning robust to complexity and length extrapolation while…
NEW PAPER ALERT: Generating visual narratives to illustrate textual stories remains an open challenge, due to the lack of knowledge to constrain faithful and self-consistent generations. Our #CVPR2025 paper proposes a new benchmark, VinaBench, to address this challenge.
📢 New paper from my internship at @cohere with @seraphinagt ‼️ Are you interested in investigating the fairness of LLMs in hiring contexts? Take a look at our work 🧵 arxiv.org/abs/2501.04316
I am officially on the job market for industry research positions focused on agentic LLMs and multi-turn reasoning! I'll be at EMNLP next week and NeurIPS next month. Message me if you'd like to chat about jobs or LLM agent research. #EMNLP2024 #neurips2024 Personal links in🧵
📣 Excited to share our latest preprint, where we investigate LLM hallucinations in multi-document summarization tasks. We reveal systematic hallucinatory behaviors across 5 popular LLMs. Check out the paper at arxiv.org/abs/2410.13961 We'd love to hear your feedback! 😄
Interested in how many images of a concept are needed for text-to-image models to imitate them, or privacy and copyright implications of text-to-image models? Check out our work led by @Sahil1V ⬇️
📣 📣 📣 Our new paper investigates the question of how many images 🖼️ of a concept are required by a diffusion model 🤖 to imitate it. This question is critical for understanding and mitigating the copyright and privacy infringements of these models! arxiv.org/abs/2410.15002
1/ 🌈Misgendering causes real harm & is often overlooked in NLP. Based on a community survey, @sunipa17, @sameer & I introduce the 1st interventions for misgendering task and eval dataset, MisgenderMender, in our #NAACL2024 paper: shorturl.at/MdzHv 🗣️: 6/18 Don Diego 9am!
Welcome to our poster at 11 am! We will introduce our #emnlp2023 work: understanding how LLMs answer multi-step reasoning questions (by memorization or step-by-step reasoning). Paper: arxiv.org/abs/2310.14491